add docker file

enforce payment finalization

.github/workflows/test.yml

@@ -59,25 +59,30 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Setup pnpm
+        uses: pnpm/action-setup@v4
+        with:
+          version: 10
+
       - name: Setup Node.js
         uses: actions/setup-node@v4
         with:
           node-version: "18"
-          cache: "npm"
-          cache-dependency-path: ui/package-lock.json
+          cache: "pnpm"
+          cache-dependency-path: ui/pnpm-lock.yaml
 
       - name: Install UI dependencies
         working-directory: ./ui
-        run: npm ci
+        run: pnpm install --frozen-lockfile
 
       - name: Run UI format check
         working-directory: ./ui
-        run: npm run format-check
+        run: pnpm run format-check
 
       - name: Run UI linting
         working-directory: ./ui
-        run: npm run lint
+        run: pnpm run lint
 
       - name: Run UI build
         working-directory: ./ui
-        run: npm run build
+        run: pnpm run build

Dockerfile.full

@@ -0,0 +1,50 @@
+# Multi-stage Dockerfile for Routstr (includes UI build)
+# Stage 1: Build the UI
+FROM node:23-alpine AS ui-builder
+WORKDIR /app/ui
+
+# Install pnpm
+RUN corepack enable pnpm && corepack prepare pnpm@latest --activate
+
+# Copy UI source
+COPY ui/package.json ui/pnpm-lock.yaml* ./
+RUN pnpm install --frozen-lockfile
+
+COPY ui/ ./
+ENV NEXT_TELEMETRY_DISABLED=1
+# Next.js build produces a static export in 'out' directory
+RUN pnpm run build
+
+# Stage 2: Build the Routstr Node
+FROM ghcr.io/astral-sh/uv:python3.11-alpine AS runner
+
+# Install system dependencies
+RUN apk add --no-cache \
+    pkgconf \
+    build-base \
+    automake \
+    autoconf \
+    libtool \
+    m4 \
+    perl \
+    git
+
+WORKDIR /app
+
+# Copy the rest of the application (required for uv sync to find the package)
+COPY . .
+
+# Install dependencies including the specific secp256k1 branch
+RUN uv add git+https://github.com/saschanaz/secp256k1-py.git#branch=upgrade060
+RUN uv sync --no-dev
+
+# Copy the built UI from the ui-builder stage
+COPY --from=ui-builder /app/ui/out ./ui_out
+
+ENV PORT=8000
+ENV PYTHONUNBUFFERED=1
+
+EXPOSE 8000
+
+# Run the application
+CMD ["/app/.venv/bin/fastapi", "run", "routstr", "--host", "0.0.0.0"]

README.md

@@ -1,256 +1,69 @@
 # Routstr Payment Proxy
 
-Routstr is a FastAPI-based reverse proxy that sits in front of any OpenAI-compatible API. It handles pay-per-request billing using the [Cashu](https://cashu.space/) eCash protocol on Bitcoin and tracks usage in a local SQL database.
+[![License](https://img.shields.io/github/license/routstr/routstr-core?style=flat-square)](LICENSE)
+[![Stars](https://img.shields.io/github/stars/routstr/routstr-core?style=flat-square)](https://github.com/routstr/routstr-core/stargazers)
+[![Issues](https://img.shields.io/github/issues/routstr/routstr-core?style=flat-square)](https://github.com/routstr/routstr-core/issues)
+[![Release](https://img.shields.io/github/v/release/routstr/routstr-core?style=flat-square)](https://github.com/routstr/routstr-core/releases)
 
-The server exposes the same endpoints as the upstream API and deducts sats from user accounts for each call. Pricing can be static or model-specific by loading `models.json` (falls back to `models.example.json`).
+Routstr is a decentralized protocol for permissionless, private, and censorship-resistant AI inference. It combines Nostr for discovery and Cashu for private Bitcoin micropayments.
 
-## How It Works
+This repo contains Routstr Core: a FastAPI-based reverse proxy that sits in front of OpenAI-compatible APIs and handles pay-per-request billing.
 
-The proxy implements a seamless eCash payment flow that maintains compatibility with existing OpenAI clients while enabling Bitcoin micropayments:
+## Start Here
 
-```mermaid
-sequenceDiagram
-    participant Client
-    participant Proxy as Routstr Proxy
-    participant DB as Database
-    participant Upstream as OpenAI API
-    participant Wallet as Cashu Wallet
+- **Overview**: <https://docs.routstr.com/overview/>
+- **Provider Guide**: <https://docs.routstr.com/provider/quickstart/>
+- **User Guide**: <https://docs.routstr.com/user-guide/introduction/>
 
-    Client->>Proxy: API Request + eCash Token
-    Proxy->>Wallet: Validate & Redeem Token
-    Wallet-->>Proxy: Token Value (sats)
-    Proxy->>DB: Store/Update Balance
-    Proxy->>Upstream: Forward API Request
-    Upstream-->>Proxy: API Response + Usage Data
-    Proxy->>DB: Deduct Actual Request Cost
-    Proxy->>DB: Update Final Balance
-    Proxy-->>Client: API Response
+## Basic Usage
+
+If you are a user/developer, you just point an OpenAI-compatible SDK at a Routstr node and pay with a Cashu token.
+
+### OpenAI SDK
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="https://api.routstr.com/v1",
+    api_key="cashuBo2FteCJodHRwczovL21...",
+)
+
+response = client.chat.completions.create(
+    model="gpt-5-nano",
+    messages=[{"role": "user", "content": "hello"}],
+)
+
+print(response.choices[0].message.content)
 ```
 
-## Features
+### cURL
 
-- **Cashu Wallet Integration** – Accept Lightning payments and redeem eCash tokens before forwarding requests
-- **API Key Management** – Hashed keys stored in SQLite with balance tracking and optional expiry/refund address
-- **Model-Based Pricing** – Convert USD prices in `models.json` to sats using live BTC/USD rates
-- **Admin Dashboard** – Simple HTML interface at `/admin/` to view balances and API keys
-- **Discovery** – Fetch available providers from Nostr relays using NIP-91 protocol
-- **NIP-91 Auto-Announcement** – Automatically announce this provider to Nostr relays when NSEC is provided
-- **Docker Support** – Provided `Dockerfile` and `compose.yml` for running with an optional Tor hidden service
+```bash
+curl https://api.routstr.com/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -H "x-cashu: cashuBo2FteCJodHRwczovL21..." \
+  -d '{
+    "model": "gpt-5-nano",
+    "messages": [{"role": "user", "content": "hello"}]
+  }'
+```
 
-## Getting Started
+## Quick Start (Docker)
 
-### Running the proxy using Docker
+If you are a node runner, start a Routstr Core instance and configure upstream access in the dashboard.
 
 ```bash
 docker run -d \
---name routstr-proxy \
--p 8000:8000 \
--e UPSTREAM_BASE_URL=https://api.openai.com/v1 \
--e UPSTREAM_API_KEY=your-openai-api-key \
-ghcr.io/routstr/proxy:latest
+  --name routstr-proxy \
+  -p 8000:8000 \
+  ghcr.io/routstr/proxy:latest
 ```
 
-### Development Requirements
-
-- Python 3.11+
-- [uv](https://github.com/astral-sh/uv) package manager (used in development)
-
-### Installation
-
-```bash
-uv sync  # install dependencies
-```
-
-Create a `.env` file based on `.env.example` and fill in the required values:
+## Development
 
 ```bash
+make setup
 cp .env.example .env
+fastapi run routstr
 ```
-
-### Running Locally
-
-```bash
-fastapi run routstr --host 0.0.0.0 --port 8000
-```
-
-The service forwards requests to `UPSTREAM_BASE_URL`. Supply the upstream API key via the `UPSTREAM_API_KEY` environment variable if required.
-
-### Docker
-
-```bash
-docker compose up --build
-```
-
-This builds the image and also starts a Tor container exposing the API as a hidden service.
-
-## Environment Variables
-
-The most common settings are shown below. See `.env.example` for the full list.
-
-### Core Settings
-
-- `UPSTREAM_BASE_URL` – URL of the OpenAI-compatible service
-- `UPSTREAM_API_KEY` – API key for the upstream service (optional)
-- `FIXED_PRICING` – Set to `true` to use a fixed per-request price; `false` (default) uses model pricing from `models.json`
-- `ADMIN_PASSWORD` – Password for the `/admin/` dashboard
-- `CASHU_MINTS` – Comma-separated list of Cashu mint URLs
-- `NAME` – Name of the proxy
-- `DESCRIPTION` – Description of the proxy
-- `NPUB` – Nostr public key of the proxy
-- `HTTP_URL` – Public-facing URL of the proxy
-- `ONION_URL` – Tor hidden service URL of the proxy
-- `NEXT_PUBLIC_API_URL` - UI Configuration for Next.js frontend (proxy URL, default: 'http://127.0.0.1:8000' )
-
-## Database Migrations
-
-The application uses Alembic for database schema management and **automatically runs migrations on startup**. This ensures your database is always up-to-date when deploying new versions.
-
-### Automatic Migrations in Production
-
-When the FastAPI application starts, it automatically:
-
-1. Runs all pending database migrations
-2. Updates the schema to the latest version
-3. Logs the migration status
-
-This means you don't need to manually run migrations when deploying - just restart the application and migrations will be applied automatically.
-
-### Manual Migration Commands
-
-For development or troubleshooting, you can use these Makefile commands:
-
-```bash
-make db-upgrade         # Apply all pending migrations
-make db-downgrade       # Downgrade one migration
-make db-current         # Show current migration revision
-make db-history         # Show migration history
-make db-migrate         # Auto-generate new migration from model changes
-make db-revision        # Create empty migration file
-make db-heads           # Show current migration heads
-make db-clean           # Clean migration cache files
-```
-
-### Creating New Migrations
-
-When you modify SQLModel models:
-
-```bash
-# Auto-generate a migration from model changes
-make db-migrate
-# Enter a descriptive message when prompted
-
-# Review the generated migration file in migrations/versions/
-# Edit if needed, then test with:
-make db-upgrade
-```
-
-## Admin UI
-
-Routstr includes a modern Next.js admin dashboard that's served directly from the Python backend as static files - no separate Node.js server required.
-
-### Building the UI
-
-```bash
-make ui-build
-```
-
-This compiles the Next.js application into static HTML, CSS, and JavaScript files in `ui/out/`.
-
-### Accessing the Dashboard
-
-Once built, the UI is automatically served by the FastAPI backend:
-
-- **Dashboard**: `http://localhost:8000/`
-- **Login**: `http://localhost:8000/login`
-- **Models Management**: `http://localhost:8000/model`
-- **Providers Management**: `http://localhost:8000/providers`
-- **Settings**: `http://localhost:8000/settings`
-
-The dashboard provides:
-
-- Real-time wallet balance monitoring
-- Model pricing configuration
-- Upstream provider management
-- Transaction history
-- System settings
-
-**Authentication**: Use the `ADMIN_PASSWORD` environment variable to access the dashboard.
-
-## Withdrawing Balance
-
-Go to the admin dashboard at `http://localhost:8000/` and login with your `ADMIN_PASSWORD` to withdraw your balance as a Cashu token.
-
-## Example Client
-
-`example.py` shows how to use the proxy with the official OpenAI client:
-
-```bash
-CASHU_TOKEN=<redeemable token> python example.py
-```
-
-The script sends streaming chat completions and pays for each request using the provided token.
-
-## Running Tests
-
-```bash
-uv run pytest
-```
-
-The tests create a temporary SQLite database and mock the Cashu wallet. See `tests/README.md` for more details.
-
-## Future Features
-
-### Nut-24 Header Support (Coming Soon)
-
-We're implementing support for the Cashu Nut-24 specification, which will enable per-request token exchange with automatic change handling:
-
-```mermaid
-graph TD
-    A["Client Request<br/>x-cashu: token"] --> B[Proxy Validates Token]
-    B --> C{Token ≥ Minimum Amount?}
-    C -->|No| F[Return 402 Payment Required]
-    C -->|Yes| D[Calculate Request Cost]
-    D --> E[Process Request]
-    E --> G[Forward to Upstream API]
-    G --> H[Receive API Response]
-    H --> I[Calculate Change]
-    I --> J["Return Response<br/>x-cashu: change_token"]
-    F --> K[End]
-    J --> K
-```
-
-**Key Benefits:**
-
-- **Per-Request Payments** – Send exact tokens for each API call
-- **Automatic Change** – Receive change tokens in response headers
-- **No Pre-funding** – No need to maintain account balances
-- **Precise Billing** – Pay only for actual usage with msat-level precision
-- **Minimum Amount Protection** – Proxy enforces minimum token value to prevent dust attacks
-
-**Header Format:**
-
-- **Request**: `x-cashu: <ecash_token>` – Token to spend for this request (must meet minimum amount)
-- **Response**: `x-cashu: <change_token>` – Change token if payment exceeds cost
-
-**Implementation Note:**
-The proxy should implement either a dedicated endpoint to communicate minimum eCash requirements per request, or extend the existing `models.json` to include minimum token amounts per model. This allows clients to autonomously determine the appropriate token amount to send with each request.
-
-**Compatible Clients:**
-
-To use this feature, you'll need a client that handles both OpenAI API calls and eCash header management. The following clients provide seamless integration:
-
-- **[routstr-chat](https://github.com/routstr/routstr-chat)** – chat app for the routstr network
-- **[otrta-client](https://github.com/routstr/otrta-client)** – rust web app for the routstr network
-
-clients automatically:
-
-- **Handle eCash Headers** – Add `x-cashu` tokens to requests and process change tokens
-- **Manage Wallets** – Maintain your Cashu wallet
-- **Configure Proxy** – Set Routstr proxy endpoints
-- **Top-up Balances** – Automatically request ecash when tokens run low and redeem ecash tokens
-
-This approach eliminates the need for account management while maintaining the security and privacy benefits of eCash payments.
-
-## License
-
-This project is licensed under the terms of the GPLv3. See the `LICENSE` file for the full license text.

compose.yml

@@ -6,7 +6,7 @@ services:
       context: ./ui
       dockerfile: Dockerfile.build
       args:
-        NEXT_PUBLIC_API_URL: ${NEXT_PUBLIC_API_URL:-http://127.0.0.1:8000}
+        # NEXT_PUBLIC_API_URL: ${NEXT_PUBLIC_API_URL:-http://127.0.0.1:8000}
         NEXT_PUBLIC_ADMIN_API_KEY: ${NEXT_PUBLIC_ADMIN_API_KEY:-}
     volumes:
       - ./ui_out:/output

docs/advanced/custom-pricing.md

@@ -1,567 +0,0 @@
-# Custom Pricing
-
-This guide covers advanced pricing strategies and customization options for Routstr Core.
-
-## Pricing Models Overview
-
-Routstr supports three pricing models:
-
-1. **Fixed Pricing** - Simple per-request fee
-2. **Token-Based Pricing** - Charge per input/output token
-3. **Model-Based Pricing** - Dynamic pricing from models.json
-
-## Model-Based Pricing
-
-### Configuration
-
-Enable model-based pricing (default behavior):
-
-```bash
-# .env
-FIXED_PRICING=false
-MODELS_PATH=/app/config/models.json
-EXCHANGE_FEE=1.005         # 0.5% exchange fee
-UPSTREAM_PROVIDER_FEE=1.05  # 5% provider margin
-```
-
-### Custom Models File
-
-Create a `models.json` with your pricing:
-
-```json
-{
-  "models": [
-    {
-      "id": "gpt-4",
-      "name": "GPT-4",
-      "description": "Advanced reasoning model",
-      "context_length": 8192,
-      "pricing": {
-        "prompt": "0.03",        // USD per 1K tokens
-        "completion": "0.06",    // USD per 1K tokens
-        "request": "0.0001",     // Fixed per-request fee
-        "image": "0",            // For multimodal models
-        "web_search": "0.005",   // Additional features
-        "internal_reasoning": "0.01"
-      },
-      "supported_features": [
-        "function_calling",
-        "vision",
-        "json_mode"
-      ],
-      "deprecation_date": null,
-      "replacement_model": null
-    },
-    {
-      "id": "custom-model",
-      "name": "Custom Fine-tuned Model",
-      "pricing": {
-        "prompt": "0.001",
-        "completion": "0.002",
-        "request": "0.00005"
-      },
-      "minimum_charge": "0.0001"  // Minimum charge per request
-    }
-  ],
-  "default_pricing": {
-    "prompt": "0.002",
-    "completion": "0.002",
-    "request": "0"
-  }
-}
-```
-
-### Dynamic Price Updates
-
-Automatically fetch prices from providers:
-
-```python
-# scripts/update_prices.py
-import asyncio
-import httpx
-import json
-
-async def fetch_openrouter_models():
-    """Fetch current model pricing from OpenRouter."""
-    async with httpx.AsyncClient() as client:
-        response = await client.get(
-            "https://openrouter.ai/api/v1/models"
-        )
-        return response.json()
-
-async def update_models_json():
-    """Update local models.json with latest prices."""
-    data = await fetch_openrouter_models()
-    
-    models = []
-    for model in data['data']:
-        models.append({
-            "id": model['id'],
-            "name": model['name'],
-            "pricing": {
-                "prompt": model['pricing']['prompt'],
-                "completion": model['pricing']['completion'],
-                "request": model['pricing'].get('request', '0')
-            },
-            "context_length": model.get('context_length', 4096)
-        })
-    
-    with open('models.json', 'w') as f:
-        json.dump({"models": models}, f, indent=2)
-
-# Run periodically
-if __name__ == "__main__":
-    asyncio.run(update_models_json())
-```
-
-## Token-Based Pricing
-
-### Configuration
-
-Set up token-based pricing overrides:
-
-```bash
-# .env
-FIXED_PRICING=false                 # use model pricing
-FIXED_COST_PER_REQUEST=1            # optional base fee
-FIXED_PER_1K_INPUT_TOKENS=5         # optional override
-FIXED_PER_1K_OUTPUT_TOKENS=15       # optional override
-```
-
-### Custom Token Counting
-
-Override default token counting:
-
-```python
-from tiktoken import encoding_for_model
-
-class CustomTokenCounter:
-    def __init__(self):
-        self.encodings = {}
-    
-    def count_tokens(
-        self, 
-        text: str, 
-        model: str
-    ) -> int:
-        """Custom token counting logic."""
-        # Cache encodings
-        if model not in self.encodings:
-            try:
-                self.encodings[model] = encoding_for_model(model)
-            except:
-                # Fallback encoding
-                self.encodings[model] = encoding_for_model("gpt-3.5-turbo")
-        
-        encoding = self.encodings[model]
-        
-        # Special handling for certain content
-        if text.startswith("```"):
-            # Code blocks might need special handling
-            tokens = encoding.encode(text)
-            return len(tokens) * 1.1  # 10% markup for code
-        
-        return len(encoding.encode(text))
-```
-
-## Advanced Pricing Strategies
-
-### Time-Based Pricing
-
-Implement peak/off-peak pricing:
-
-```python
-from datetime import datetime
-import pytz
-
-class TimeBased PricingStrategy:
-    def __init__(self):
-        self.timezone = pytz.timezone('US/Eastern')
-        self.peak_hours = [(9, 17)]  # 9 AM - 5 PM
-        self.peak_multiplier = 1.5
-        self.weekend_discount = 0.8
-    
-    def get_price_multiplier(self) -> float:
-        """Calculate price multiplier based on time."""
-        now = datetime.now(self.timezone)
-        
-        # Weekend discount
-        if now.weekday() >= 5:  # Saturday or Sunday
-            return self.weekend_discount
-        
-        # Peak hours surcharge
-        hour = now.hour
-        for start, end in self.peak_hours:
-            if start <= hour < end:
-                return self.peak_multiplier
-        
-        # Off-peak standard pricing
-        return 1.0
-    
-    def apply_to_cost(self, base_cost: int) -> int:
-        """Apply time-based pricing to cost."""
-        multiplier = self.get_price_multiplier()
-        return int(base_cost * multiplier)
-```
-
-### Model-Specific Surcharges
-
-Add custom fees for specific models:
-
-```python
-class ModelSurchargeStrategy:
-    def __init__(self):
-        self.surcharges = {
-            "gpt-4-turbo": 1.1,      # 10% premium
-            "claude-3-opus": 1.15,    # 15% premium
-            "dall-e-3-hd": 1.25,      # 25% premium for HD
-        }
-        
-        self.discounts = {
-            "gpt-3.5-turbo": 0.95,    # 5% discount
-            "deprecated-model": 0.8,   # 20% discount
-        }
-    
-    def get_model_multiplier(self, model: str) -> float:
-        """Get price multiplier for model."""
-        if model in self.surcharges:
-            return self.surcharges[model]
-        elif model in self.discounts:
-            return self.discounts[model]
-        return 1.0
-```
-
-### Geographic Pricing
-
-Adjust pricing based on client location:
-
-```python
-import geoip2.database
-
-class GeographicPricingStrategy:
-    def __init__(self):
-        self.reader = geoip2.database.Reader('GeoLite2-Country.mmdb')
-        self.country_multipliers = {
-            'US': 1.0,
-            'GB': 1.0,
-            'DE': 1.0,
-            'IN': 0.7,  # 30% discount
-            'BR': 0.8,  # 20% discount
-            'NG': 0.6,  # 40% discount
-        }
-        self.default_multiplier = 0.9
-    
-    def get_country_multiplier(self, ip_address: str) -> float:
-        """Get price multiplier based on country."""
-        try:
-            response = self.reader.country(ip_address)
-            country_code = response.country.iso_code
-            return self.country_multipliers.get(
-                country_code, 
-                self.default_multiplier
-            )
-        except:
-            return 1.0  # Default pricing if lookup fails
-```
-
-## Cost Calculation Pipeline
-
-### Implementing Custom Calculator
-
-```python
-from abc import ABC, abstractmethod
-
-class CostCalculator(ABC):
-    @abstractmethod
-    async def calculate(
-        self,
-        request_data: dict,
-        usage_data: dict,
-        context: dict
-    ) -> CostResult:
-        pass
-
-class CompositeCostCalculator(CostCalculator):
-    """Combine multiple pricing strategies."""
-    
-    def __init__(self):
-        self.strategies = [
-            BaseCostCalculator(),
-            TimeBasedPricingStrategy(),
-            ModelSurchargeStrategy(),
-            GeographicPricingStrategy()
-        ]
-    
-    async def calculate(
-        self,
-        request_data: dict,
-        usage_data: dict,
-        context: dict
-    ) -> CostResult:
-        # Start with base cost
-        base_cost = await self.strategies[0].calculate(
-            request_data, usage_data, context
-        )
-        
-        # Apply each strategy
-        final_cost = base_cost.total_msats
-        breakdown = {"base": base_cost.total_msats}
-        
-        for strategy in self.strategies[1:]:
-            multiplier = await strategy.get_multiplier(context)
-            adjustment = final_cost * (multiplier - 1)
-            final_cost += adjustment
-            breakdown[strategy.__class__.__name__] = adjustment
-        
-        return CostResult(
-            total_msats=int(final_cost),
-            breakdown=breakdown
-        )
-```
-
-### Integration with Routstr
-
-```python
-# In routstr/payment/cost_calculation.py
-async def calculate_request_cost(
-    model: str,
-    prompt_tokens: int,
-    completion_tokens: int,
-    request_type: str,
-    context: dict
-) -> CostData:
-    """Enhanced cost calculation with custom strategies."""
-    
-    # Use custom calculator if configured
-    if os.getenv("USE_CUSTOM_PRICING", "false").lower() == "true":
-        calculator = CompositeCostCalculator()
-        result = await calculator.calculate(
-            request_data={
-                "model": model,
-                "type": request_type
-            },
-            usage_data={
-                "prompt_tokens": prompt_tokens,
-                "completion_tokens": completion_tokens
-            },
-            context=context
-        )
-        return result
-    
-    # Fall back to standard calculation
-    return standard_calculate_cost(...)
-```
-
-## Monitoring Pricing
-
-### Price Analytics
-
-Track pricing effectiveness:
-
-```python
-class PricingAnalytics:
-    async def analyze_pricing(
-        self,
-        start_date: datetime,
-        end_date: datetime
-    ):
-        """Analyze pricing performance."""
-        # Average cost per request by model
-        model_costs = await self.get_average_costs_by_model(
-            start_date, end_date
-        )
-        
-        # Revenue by pricing strategy
-        strategy_revenue = await self.get_revenue_by_strategy(
-            start_date, end_date
-        )
-        
-        # Price elasticity
-        elasticity = await self.calculate_price_elasticity()
-        
-        return {
-            "model_costs": model_costs,
-            "strategy_revenue": strategy_revenue,
-            "price_elasticity": elasticity,
-            "recommendations": self.generate_recommendations(
-                model_costs, elasticity
-            )
-        }
-```
-
-### A/B Testing Prices
-
-Test different pricing strategies:
-
-```python
-class PricingExperiment:
-    def __init__(self):
-        self.experiments = {
-            "exp_001": {
-                "name": "10% discount test",
-                "group_a": {"multiplier": 1.0},
-                "group_b": {"multiplier": 0.9},
-                "allocation": 0.5  # 50/50 split
-            }
-        }
-    
-    def assign_group(self, api_key_id: int) -> str:
-        """Assign API key to experiment group."""
-        # Consistent assignment based on key ID
-        import hashlib
-        hash_value = int(hashlib.md5(
-            str(api_key_id).encode()
-        ).hexdigest()[:8], 16)
-        
-        return "group_b" if (hash_value % 100) < 50 else "group_a"
-    
-    def get_experiment_multiplier(
-        self, 
-        api_key_id: int,
-        experiment_id: str
-    ) -> float:
-        """Get price multiplier for experiment."""
-        experiment = self.experiments.get(experiment_id)
-        if not experiment:
-            return 1.0
-        
-        group = self.assign_group(api_key_id)
-        return experiment[group]["multiplier"]
-```
-
-## Configuration Examples
-
-### Enterprise Pricing
-
-```json
-{
-  "models": [
-    {
-      "id": "gpt-4-enterprise",
-      "name": "GPT-4 Enterprise",
-      "pricing": {
-        "prompt": "0.02",
-        "completion": "0.04"
-      },
-      "minimum_commitment": "1000",  // $1000/month minimum
-      "sla": {
-        "uptime": "99.9%",
-        "support_response": "1 hour",
-        "dedicated_capacity": true
-      }
-    }
-  ],
-  "enterprise_features": {
-    "priority_queue": true,
-    "custom_models": true,
-    "audit_logs": true,
-    "sso": true
-  }
-}
-```
-
-### Budget-Friendly Options
-
-```json
-{
-  "models": [
-    {
-      "id": "gpt-3.5-turbo-budget",
-      "name": "GPT-3.5 Turbo Budget",
-      "pricing": {
-        "prompt": "0.0005",
-        "completion": "0.001"
-      },
-      "restrictions": {
-        "max_tokens_per_request": 1000,
-        "requests_per_minute": 10,
-        "peak_hours_blocked": true
-      }
-    }
-  ],
-  "prepaid_packages": [
-    {
-      "name": "Starter Pack",
-      "price_usd": 10,
-      "tokens_included": 10000000,
-      "expires_days": 30
-    }
-  ]
-}
-```
-
-## Troubleshooting
-
-### Price Calculation Issues
-
-```python
-# Debug pricing
-async def debug_price_calculation(
-    model: str,
-    tokens: dict,
-    api_key_id: int
-):
-    """Debug price calculation step by step."""
-    print(f"Model: {model}")
-    print(f"Tokens: {tokens}")
-    
-    # Base price
-    base_price = get_model_price(model)
-    print(f"Base price: {base_price}")
-    
-    # Token cost
-    token_cost = calculate_token_cost(base_price, tokens)
-    print(f"Token cost: {token_cost}")
-    
-    # Strategies
-    strategies = get_active_strategies()
-    for strategy in strategies:
-        multiplier = await strategy.get_multiplier(api_key_id)
-        print(f"{strategy.name}: {multiplier}x")
-    
-    # Final cost
-    final_cost = apply_all_strategies(token_cost, api_key_id)
-    print(f"Final cost: {final_cost} msats")
-    
-    return final_cost
-```
-
-### Common Issues
-
-1. **Prices Not Updating**
-   - Check `MODELS_PATH` is correct
-   - Verify file permissions
-   - Check background task logs
-
-2. **Wrong Currency Conversion**
-   - Verify BTC/USD rate source
-   - Check `EXCHANGE_FEE` setting
-   - Monitor rate update frequency
-
-3. **Discounts Not Applied**
-   - Verify strategy configuration
-   - Check API key metadata
-   - Review transaction history
-
-## Best Practices
-
-1. **Transparent Pricing**
-   - Publish pricing clearly
-   - Show cost breakdowns
-   - Notify of price changes
-
-2. **Fair Pricing**
-   - Regular competitive analysis
-   - Consider user feedback
-   - Offer budget options
-
-3. **Performance**
-   - Cache price calculations
-   - Optimize database queries
-   - Monitor calculation time
-
-## Next Steps
-
-- [Migrations](migrations.md) - Database migration guide
-- [API Endpoints](../api/endpoints.md) - Pricing endpoints
-- [Monitoring](../user-guide/admin-dashboard.md) - Track pricing metrics

docs/advanced/migrations.md

@@ -1,646 +0,0 @@
-# Database Migrations
-
-This guide covers database schema management using Alembic migrations in Routstr Core.
-
-## Overview
-
-Routstr uses Alembic for database migrations with these features:
-
-- **Automatic migrations** on startup
-- **Version control** for schema changes
-- **Rollback capability** for safety
-- **Support for multiple databases** (SQLite, PostgreSQL)
-
-## Automatic Migrations
-
-### Startup Behavior
-
-Migrations run automatically when Routstr starts:
-
-```python
-# In routstr/core/main.py
-@asynccontextmanager
-async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
-    logger.info("Running database migrations")
-    run_migrations()  # Automatic migration
-    await init_db()   # Initialize connection pool
-    # ... rest of startup
-```
-
-This ensures:
-
-- ✅ Database is always up-to-date
-- ✅ No manual migration steps in production
-- ✅ Zero-downtime deployments
-- ✅ Backwards compatibility
-
-### Migration Safety
-
-Migrations are designed to be safe:
-
-- Idempotent (can run multiple times)
-- Non-destructive by default
-- Tested before release
-- Reversible when possible
-
-## Creating Migrations
-
-### Auto-generating from Models
-
-After modifying SQLModel classes:
-
-```bash
-# Generate migration from model changes
-make db-migrate
-
-# You'll be prompted for a description
-Enter migration message: Add user preferences table
-
-# Review generated file
-cat migrations/versions/xxxx_add_user_preferences_table.py
-```
-
-### Manual Migrations
-
-For complex changes, create manually:
-
-```bash
-# Create empty migration
-alembic revision -m "Complex data transformation"
-
-# Edit the generated file
-vim migrations/versions/xxxx_complex_data_transformation.py
-```
-
-### Migration Template
-
-```python
-"""Add user preferences table
-
-Revision ID: a1b2c3d4e5f6
-Revises: f6e5d4c3b2a1
-Create Date: 2024-01-15 10:30:00.123456
-
-"""
-from alembic import op
-import sqlalchemy as sa
-import sqlmodel
-
-# revision identifiers
-revision = 'a1b2c3d4e5f6'
-down_revision = 'f6e5d4c3b2a1'
-branch_labels = None
-depends_on = None
-
-def upgrade() -> None:
-    """Apply migration."""
-    # Create new table
-    op.create_table(
-        'userpreferences',
-        sa.Column('id', sa.Integer(), nullable=False),
-        sa.Column('api_key_id', sa.Integer(), nullable=False),
-        sa.Column('theme', sa.String(), nullable=True),
-        sa.Column('notifications_enabled', sa.Boolean(), default=True),
-        sa.Column('created_at', sa.DateTime(), nullable=False),
-        sa.PrimaryKeyConstraint('id'),
-        sa.ForeignKeyConstraint(['api_key_id'], ['apikey.id'], )
-    )
-    
-    # Create index
-    op.create_index(
-        'ix_userpreferences_api_key_id',
-        'userpreferences',
-        ['api_key_id']
-    )
-
-def downgrade() -> None:
-    """Revert migration."""
-    op.drop_index('ix_userpreferences_api_key_id', table_name='userpreferences')
-    op.drop_table('userpreferences')
-```
-
-## Common Migration Patterns
-
-### Adding Columns
-
-Add column with default value:
-
-```python
-def upgrade():
-    # Add nullable column first
-    op.add_column(
-        'apikey',
-        sa.Column('last_rotation', sa.DateTime(), nullable=True)
-    )
-    
-    # Populate existing rows
-    connection = op.get_bind()
-    connection.execute(
-        "UPDATE apikey SET last_rotation = created_at WHERE last_rotation IS NULL"
-    )
-    
-    # Make non-nullable if needed
-    op.alter_column('apikey', 'last_rotation', nullable=False)
-```
-
-### Renaming Columns
-
-Safe column rename:
-
-```python
-def upgrade():
-    # SQLite doesn't support ALTER COLUMN, so we need a workaround
-    with op.batch_alter_table('apikey') as batch_op:
-        batch_op.alter_column('old_name', new_column_name='new_name')
-```
-
-### Adding Indexes
-
-Performance-improving indexes:
-
-```python
-def upgrade():
-    # Single column index
-    op.create_index(
-        'ix_transaction_timestamp',
-        'transaction',
-        ['timestamp']
-    )
-    
-    # Composite index
-    op.create_index(
-        'ix_transaction_key_time',
-        'transaction',
-        ['api_key_id', 'timestamp']
-    )
-    
-    # Partial index (PostgreSQL only)
-    op.create_index(
-        'ix_apikey_active',
-        'apikey',
-        ['balance'],
-        postgresql_where='balance > 0'
-    )
-```
-
-### Data Migrations
-
-Transform existing data:
-
-```python
-def upgrade():
-    # Add new column
-    op.add_column(
-        'apikey',
-        sa.Column('key_type', sa.String(), nullable=True)
-    )
-    
-    # Migrate data
-    connection = op.get_bind()
-    result = connection.execute('SELECT id, metadata FROM apikey')
-    
-    for row in result:
-        key_type = 'premium' if row.metadata.get('premium') else 'standard'
-        connection.execute(
-            f"UPDATE apikey SET key_type = '{key_type}' WHERE id = {row.id}"
-        )
-    
-    # Make column non-nullable
-    op.alter_column('apikey', 'key_type', nullable=False)
-```
-
-### Enum Types
-
-Add enum column:
-
-```python
-from enum import Enum
-
-class KeyStatus(str, Enum):
-    ACTIVE = "active"
-    SUSPENDED = "suspended"
-    EXPIRED = "expired"
-
-def upgrade():
-    # Create enum type (PostgreSQL)
-    key_status_enum = sa.Enum(KeyStatus, name='keystatus')
-    key_status_enum.create(op.get_bind(), checkfirst=True)
-    
-    # Add column
-    op.add_column(
-        'apikey',
-        sa.Column(
-            'status',
-            key_status_enum,
-            nullable=False,
-            server_default='active'
-        )
-    )
-```
-
-## Database-Specific Considerations
-
-### SQLite Limitations
-
-SQLite has limitations requiring workarounds:
-
-```python
-def upgrade():
-    # SQLite doesn't support ALTER COLUMN directly
-    # Use batch_alter_table for compatibility
-    with op.batch_alter_table('apikey') as batch_op:
-        batch_op.alter_column(
-            'balance',
-            type_=sa.BigInteger(),  # Change from Integer
-            existing_type=sa.Integer()
-        )
-```
-
-### PostgreSQL Features
-
-Leverage PostgreSQL-specific features:
-
-```python
-def upgrade():
-    # Use JSONB for better performance
-    op.add_column(
-        'apikey',
-        sa.Column('metadata', sa.JSON().with_variant(
-            sa.dialects.postgresql.JSONB(), 'postgresql'
-        ))
-    )
-    
-    # Add GIN index for JSONB queries
-    op.create_index(
-        'ix_apikey_metadata',
-        'apikey',
-        ['metadata'],
-        postgresql_using='gin'
-    )
-    
-    # Add check constraint
-    op.create_check_constraint(
-        'ck_apikey_balance_positive',
-        'apikey',
-        'balance >= 0'
-    )
-```
-
-## Migration Commands
-
-### Running Migrations
-
-```bash
-# Apply all pending migrations
-make db-upgrade
-
-# Upgrade to specific revision
-alembic upgrade a1b2c3d4e5f6
-
-# Upgrade one revision
-alembic upgrade +1
-```
-
-### Checking Status
-
-```bash
-# Show current revision
-make db-current
-# Output: a1b2c3d4e5f6 (head)
-
-# Show migration history
-make db-history
-# Output:
-# a1b2c3d4e5f6 -> b2c3d4e5f6a7 (head), Add user preferences
-# f6e5d4c3b2a1 -> a1b2c3d4e5f6, Add indexes
-# e5d4c3b2a1f6 -> f6e5d4c3b2a1, Initial schema
-```
-
-### Rolling Back
-
-```bash
-# Rollback one migration
-make db-downgrade
-
-# Rollback to specific revision
-alembic downgrade f6e5d4c3b2a1
-
-# Rollback all (dangerous!)
-alembic downgrade base
-```
-
-## Testing Migrations
-
-### Unit Testing
-
-Test migrations in isolation:
-
-```python
-import pytest
-from alembic import command
-from alembic.config import Config
-from sqlalchemy import create_engine, inspect
-
-def test_migration_add_user_preferences():
-    """Test user preferences migration."""
-    # Create test database
-    engine = create_engine("sqlite:///:memory:")
-    
-    # Run migrations up to previous version
-    alembic_cfg = Config("alembic.ini")
-    alembic_cfg.set_main_option("sqlalchemy.url", str(engine.url))
-    command.upgrade(alembic_cfg, "f6e5d4c3b2a1")
-    
-    # Verify state before migration
-    inspector = inspect(engine)
-    tables = inspector.get_table_names()
-    assert "userpreferences" not in tables
-    
-    # Run target migration
-    command.upgrade(alembic_cfg, "a1b2c3d4e5f6")
-    
-    # Verify state after migration
-    inspector = inspect(engine)
-    tables = inspector.get_table_names()
-    assert "userpreferences" in tables
-    
-    # Check columns
-    columns = {col['name'] for col in inspector.get_columns('userpreferences')}
-    assert columns == {'id', 'api_key_id', 'theme', 'notifications_enabled', 'created_at'}
-    
-    # Test downgrade
-    command.downgrade(alembic_cfg, "f6e5d4c3b2a1")
-    inspector = inspect(engine)
-    tables = inspector.get_table_names()
-    assert "userpreferences" not in tables
-```
-
-### Integration Testing
-
-Test with real data:
-
-```python
-async def test_migration_with_data():
-    """Test migration preserves existing data."""
-    # Setup test database with data
-    async with test_engine.begin() as conn:
-        # Insert test data
-        await conn.execute(
-            "INSERT INTO apikey (key_hash, balance) VALUES ('test', 1000)"
-        )
-    
-    # Run migration
-    run_migrations()
-    
-    # Verify data integrity
-    async with test_engine.connect() as conn:
-        result = await conn.execute("SELECT * FROM apikey WHERE key_hash = 'test'")
-        row = result.first()
-        assert row.balance == 1000
-        assert row.key_type == 'standard'  # New column with default
-```
-
-## Production Deployment
-
-### Zero-Downtime Migrations
-
-Strategy for seamless updates:
-
-1. **Make migrations backwards compatible**
-
-   ```python
-   # Good: Add nullable column
-   op.add_column('apikey', sa.Column('new_field', sa.String(), nullable=True))
-   
-   # Bad: Drop column immediately
-   # op.drop_column('apikey', 'old_field')
-   ```
-
-2. **Deploy in phases**
-
-   ```bash
-   # Phase 1: Deploy code that works with both schemas
-   # Phase 2: Run migration
-   # Phase 3: Deploy code that requires new schema
-   # Phase 4: Clean up deprecated columns
-   ```
-
-3. **Use feature flags**
-
-   ```python
-   if feature_enabled('use_new_schema'):
-       # Use new column
-       query = select(APIKey.new_field)
-   else:
-       # Use old column
-       query = select(APIKey.old_field)
-   ```
-
-### Migration Monitoring
-
-Track migration execution:
-
-```python
-# Add to migration
-def upgrade():
-    start_time = time.time()
-    logger.info(f"Starting migration {revision}")
-    
-    try:
-        # Migration logic here
-        op.create_table(...)
-        
-        duration = time.time() - start_time
-        logger.info(f"Migration {revision} completed in {duration:.2f}s")
-    except Exception as e:
-        logger.error(f"Migration {revision} failed: {e}")
-        raise
-```
-
-### Backup Before Migration
-
-Always backup before major changes:
-
-```bash
-#!/bin/bash
-# backup_before_migration.sh
-
-# Backup database
-if [[ "$DATABASE_URL" == *"sqlite"* ]]; then
-    cp database.db "backup_$(date +%Y%m%d_%H%M%S).db"
-else
-    pg_dump $DATABASE_URL > "backup_$(date +%Y%m%d_%H%M%S).sql"
-fi
-
-# Run migration
-alembic upgrade head
-
-# Verify
-alembic current
-```
-
-## Troubleshooting
-
-### Common Issues
-
-**Migration Conflicts**
-
-```bash
-# Multiple heads detected
-alembic heads
-# a1b2c3d4e5f6 (head)
-# b2c3d4e5f6a7 (head)
-
-# Merge heads
-alembic merge -m "Merge migrations" a1b2c3 b2c3d4
-```
-
-**Failed Migration**
-
-```python
-# Add rollback logic
-def upgrade():
-    try:
-        op.create_table(...)
-    except Exception as e:
-        # Clean up partial changes
-        op.drop_table('partial_table', checkfirst=True)
-        raise
-
-def downgrade():
-    # Ensure clean rollback
-    op.drop_table('new_table', checkfirst=True)
-```
-
-**Lock Timeout**
-
-```python
-# Add timeout handling
-def upgrade():
-    connection = op.get_bind()
-    
-    # Set timeout (PostgreSQL)
-    connection.execute("SET lock_timeout = '10s'")
-    
-    try:
-        op.add_column(...)
-    except OperationalError as e:
-        if 'lock timeout' in str(e):
-            logger.error("Migration failed due to lock timeout")
-            raise
-```
-
-### Recovery Procedures
-
-If migration fails in production:
-
-1. **Check current state**
-
-   ```bash
-   alembic current
-   alembic history
-   ```
-
-2. **Manual rollback if needed**
-
-   ```sql
-   -- Check migration table
-   SELECT * FROM alembic_version;
-   
-   -- Force version if necessary
-   UPDATE alembic_version SET version_num = 'previous_version';
-   ```
-
-3. **Fix and retry**
-
-   ```bash
-   # Fix migration file
-   vim migrations/versions/problematic_migration.py
-   
-   # Retry
-   alembic upgrade head
-   ```
-
-## Best Practices
-
-### Migration Guidelines
-
-1. **Keep migrations small and focused**
-   - One logical change per migration
-   - Easier to review and rollback
-
-2. **Test migrations thoroughly**
-   - Test upgrade and downgrade
-   - Test with production-like data
-   - Test database-specific features
-
-3. **Document breaking changes**
-
-   ```python
-   """BREAKING: Change balance column type
-   
-   This migration requires application update.
-   Deploy order:
-   1. Update application to handle both int and bigint
-   2. Run this migration
-   3. Update application to use only bigint
-   """
-   ```
-
-4. **Make migrations idempotent**
-
-   ```python
-   def upgrade():
-       # Check if column exists
-       inspector = inspect(op.get_bind())
-       columns = [col['name'] for col in inspector.get_columns('apikey')]
-       
-       if 'new_column' not in columns:
-           op.add_column(
-               'apikey',
-               sa.Column('new_column', sa.String())
-           )
-   ```
-
-### Performance Considerations
-
-1. **Add indexes concurrently (PostgreSQL)**
-
-   ```python
-   def upgrade():
-       # Create index without locking table
-       op.create_index(
-           'ix_large_table_column',
-           'large_table',
-           ['column'],
-           postgresql_concurrently=True
-       )
-   ```
-
-2. **Batch large updates**
-
-   ```python
-   def upgrade():
-       connection = op.get_bind()
-       
-       # Process in batches
-       batch_size = 1000
-       offset = 0
-       
-       while True:
-           result = connection.execute(
-               f"UPDATE apikey SET processed = true "
-               f"WHERE id IN (SELECT id FROM apikey WHERE processed = false LIMIT {batch_size})"
-           )
-           
-           if result.rowcount == 0:
-               break
-               
-           offset += batch_size
-           time.sleep(0.1)  # Prevent overload
-   ```
-
-## Next Steps
-
-- [Testing Guide](../contributing/testing.md) - Testing migrations
-- [Deployment](../getting-started/docker.md) - Production deployment

docs/advanced/nostr.md

@@ -1,598 +0,0 @@
-# Nostr Discovery
-
-Routstr Core integrates with Nostr (Notes and Other Stuff Transmitted by Relays) for decentralized provider discovery. This enables users to find Routstr nodes without relying on centralized directories.
-
-## Overview
-
-Nostr integration provides:
-
-- **Decentralized Discovery**: Find providers through relay network
-- **Cryptographic Identity**: Providers identified by public keys
-- **Real-time Updates**: Live provider status and pricing
-- **Censorship Resistance**: No central point of control
-
-## How It Works
-
-```mermaid
-graph LR
-    A[Routstr Node] --> B[Nostr Relay]
-    B --> C[Nostr Relay]
-    B --> D[Nostr Relay]
-    
-    E[User Client] --> B
-    E --> C
-    E --> D
-    
-    B --> F[Provider List]
-    C --> F
-    D --> F
-```
-
-Providers announce themselves by publishing signed events to Nostr relays. Clients can query these relays to discover available providers.
-
-## Provider Configuration
-
-### Setting Up Nostr Identity
-
-1. **Generate Nostr Keys**
-
-   ```bash
-   # Using nostril or similar tool
-   nostril --generate-keypair
-   
-   # Output:
-   # Private key (nsec): nsec1abc...
-   # Public key (npub): npub1xyz...
-   ```
-
-2. **Configure Environment**
-
-   ```bash
-   # .env
-   NPUB=npub1xyz...  # Your public key
-   NSEC=nsec1abc...  # Your private key (keep secret!)
-   NAME=Lightning AI Gateway
-   DESCRIPTION=Fast and reliable AI API with Bitcoin payments
-   HTTP_URL=https://api.lightning-ai.com
-   ONION_URL=http://lightningai.onion
-   ```
-
-### Publishing to Nostr
-
-Routstr automatically publishes provider information to configured relays:
-
-```python
-# Published event structure (NIP-89)
-{
-    "kind": 31990,  # Application handler event
-    "pubkey": "your_public_key",
-    "content": {
-        "name": "Lightning AI Gateway",
-        "description": "Fast and reliable AI API",
-        "endpoints": {
-            "http": "https://api.lightning-ai.com",
-            "onion": "http://lightningai.onion"
-        },
-        "models": ["gpt-3.5-turbo", "gpt-4", "claude-3"],
-        "pricing": {
-            "gpt-3.5-turbo": {
-                "prompt_sats_per_1k": 3,
-                "completion_sats_per_1k": 4
-            }
-        },
-        "cashu_mints": [
-            "https://mint.minibits.cash/Bitcoin"
-        ]
-    },
-    "tags": [
-        ["d", "routstr"],
-        ["t", "ai-api"],
-        ["t", "bitcoin"],
-        ["p", "payment-proxy"]
-    ]
-}
-```
-
-### Relay Configuration
-
-Configure which relays to publish to:
-
-```python
-# Default relays
-DEFAULT_RELAYS = [
-    "wss://relay.damus.io",
-    "wss://relay.nostr.band",
-    "wss://relay.routstr.com",
-    "wss://nos.lol"
-]
-
-# Custom relay configuration
-RELAYS=wss://relay1.com,wss://relay2.com
-```
-
-## Client Discovery
-
-### Using the Discovery Endpoint
-
-Find providers through the API:
-
-```bash
-GET /v1/providers
-
-Response:
-{
-  "providers": [
-    {
-      "name": "Lightning AI Gateway",
-      "npub": "npub1xyz...",
-      "description": "Fast and reliable AI API",
-      "endpoints": {
-        "http": "https://api.lightning-ai.com",
-        "onion": "http://lightningai.onion"
-      },
-      "models": ["gpt-3.5-turbo", "gpt-4"],
-      "pricing": {
-        "gpt-3.5-turbo": {
-          "prompt_sats_per_1k": 3,
-          "completion_sats_per_1k": 4
-        }
-      },
-      "last_seen": "2024-01-01T12:00:00Z",
-      "reliability_score": 0.99
-    }
-  ]
-}
-```
-
-### Direct Nostr Queries
-
-Query Nostr relays directly:
-
-```python
-import json
-import websocket
-
-def discover_providers(relay_url: str):
-    """Discover Routstr providers from Nostr relay."""
-    ws = websocket.create_connection(relay_url)
-    
-    # Subscribe to provider events
-    subscription = {
-        "kinds": [31990],
-        "tags": {
-            "d": ["routstr"]
-        }
-    }
-    
-    ws.send(json.dumps(["REQ", "sub1", subscription]))
-    
-    providers = []
-    while True:
-        response = json.loads(ws.recv())
-        if response[0] == "EVENT":
-            event = response[2]
-            providers.append(parse_provider_event(event))
-        elif response[0] == "EOSE":  # End of stored events
-            break
-    
-    ws.close()
-    return providers
-```
-
-### JavaScript/TypeScript
-
-```typescript
-import { SimplePool } from 'nostr-tools';
-
-async function discoverProviders(): Promise<Provider[]> {
-    const pool = new SimplePool();
-    const relays = [
-        'wss://relay.damus.io',
-        'wss://relay.nostr.band'
-    ];
-    
-    const filter = {
-        kinds: [31990],
-        '#d': ['routstr']
-    };
-    
-    const events = await pool.list(relays, [filter]);
-    
-    return events.map(event => ({
-        name: event.content.name,
-        npub: nip19.npubEncode(event.pubkey),
-        url: event.content.endpoints.http,
-        models: event.content.models,
-        pricing: event.content.pricing
-    }));
-}
-```
-
-## Provider Ranking
-
-### Reliability Scoring
-
-Providers are ranked based on:
-
-```python
-class ProviderScore:
-    def calculate(self, provider: Provider) -> float:
-        score = 1.0
-        
-        # Uptime (based on recent checks)
-        uptime_ratio = provider.successful_pings / provider.total_pings
-        score *= uptime_ratio
-        
-        # Response time
-        if provider.avg_response_time < 500:  # ms
-            score *= 1.0
-        elif provider.avg_response_time < 1000:
-            score *= 0.9
-        else:
-            score *= 0.7
-        
-        # Model availability
-        model_score = len(provider.models) / 10  # Max 10 models
-        score *= min(1.0, 0.5 + model_score * 0.5)
-        
-        # Price competitiveness
-        if provider.is_cheapest_for_any_model():
-            score *= 1.1
-        
-        return min(1.0, score)
-```
-
-### Provider Selection
-
-Choose optimal provider:
-
-```python
-def select_provider(
-    providers: list[Provider],
-    model: str,
-    requirements: dict
-) -> Provider:
-    """Select best provider for requirements."""
-    
-    # Filter by model availability
-    candidates = [p for p in providers if model in p.models]
-    
-    # Filter by requirements
-    if requirements.get('tor_required'):
-        candidates = [p for p in candidates if p.onion_url]
-    
-    if requirements.get('max_price_per_1k'):
-        max_price = requirements['max_price_per_1k']
-        candidates = [
-            p for p in candidates
-            if p.pricing[model]['prompt_sats_per_1k'] <= max_price
-        ]
-    
-    # Sort by score
-    candidates.sort(key=lambda p: p.reliability_score, reverse=True)
-    
-    return candidates[0] if candidates else None
-```
-
-## Publishing Updates
-
-### Automatic Updates
-
-Routstr publishes updates when:
-
-- Node starts up
-- Configuration changes
-- Models are added/removed
-- Pricing updates
-
-### Manual Publishing
-
-Force publish current state:
-
-```python
-async def publish_provider_info():
-    """Manually publish provider information."""
-    event = create_provider_event(
-        name=os.getenv("NAME"),
-        description=os.getenv("DESCRIPTION"),
-        models=get_available_models(),
-        pricing=get_current_pricing()
-    )
-    
-    await publish_to_relays(event, RELAYS)
-```
-
-### Event Lifecycle
-
-```python
-# Publish every 6 hours
-@periodic_task(hours=6)
-async def update_nostr_presence():
-    """Keep provider information fresh."""
-    try:
-        await publish_provider_info()
-        logger.info("Updated Nostr presence")
-    except Exception as e:
-        logger.error(f"Failed to update Nostr: {e}")
-
-# Delete on shutdown
-async def remove_nostr_presence():
-    """Remove provider from discovery."""
-    deletion_event = create_deletion_event()
-    await publish_to_relays(deletion_event, RELAYS)
-```
-
-## Security Considerations
-
-### Key Management
-
-1. **Secure Storage**
-
-   ```python
-   # Never log private keys
-   SENSITIVE_VARS = ['NSEC', 'ADMIN_PASSWORD']
-   
-   def sanitize_env(env_dict: dict) -> dict:
-       return {
-           k: '***' if k in SENSITIVE_VARS else v
-           for k, v in env_dict.items()
-       }
-   ```
-
-2. **Key Rotation**
-
-   ```bash
-   # Generate new keys
-   nostril --generate-keypair
-   
-   # Update configuration
-   # Publish transition event
-   # Update all references
-   ```
-
-### Event Validation
-
-Verify provider events:
-
-```python
-def validate_provider_event(event: dict) -> bool:
-    """Validate provider announcement."""
-    # Check signature
-    if not verify_signature(event):
-        return False
-    
-    # Check required fields
-    required = ['name', 'endpoints', 'models', 'pricing']
-    content = json.loads(event['content'])
-    if not all(field in content for field in required):
-        return False
-    
-    # Verify endpoints are reachable
-    if not await check_endpoints(content['endpoints']):
-        return False
-    
-    return True
-```
-
-### Relay Security
-
-Choose relays carefully:
-
-```python
-TRUSTED_RELAYS = {
-    'wss://relay.damus.io': {
-        'operator': 'Damus',
-        'reputation': 'high',
-        'filters_spam': True
-    },
-    'wss://relay.nostr.band': {
-        'operator': 'Nostr.Band',
-        'reputation': 'high',
-        'paid_tier': True
-    }
-}
-```
-
-## Advanced Features
-
-### Multi-Relay Broadcasting
-
-Ensure wide distribution:
-
-```python
-async def broadcast_to_relays(event: dict, relays: list[str]):
-    """Broadcast event to multiple relays."""
-    tasks = []
-    for relay in relays:
-        task = asyncio.create_task(
-            publish_to_relay(event, relay)
-        )
-        tasks.append(task)
-    
-    results = await asyncio.gather(*tasks, return_exceptions=True)
-    
-    successful = sum(1 for r in results if not isinstance(r, Exception))
-    logger.info(f"Published to {successful}/{len(relays)} relays")
-```
-
-### Provider Metadata
-
-Extended metadata in events:
-
-```json
-{
-    "kind": 31990,
-    "content": {
-        "name": "Lightning AI",
-        "description": "Enterprise AI API",
-        "metadata": {
-            "established": "2024-01-01",
-            "total_requests": 1000000,
-            "average_response_ms": 250,
-            "supported_features": [
-                "streaming",
-                "function_calling",
-                "vision",
-                "embeddings"
-            ],
-            "certifications": ["SOC2", "GDPR"],
-            "contact": {
-                "nostr": "npub1contact...",
-                "email": "support@lightning-ai.com"
-            }
-        }
-    }
-}
-```
-
-### Discovery Filters
-
-Advanced filtering options:
-
-```python
-# Find providers with specific features
-GET /v1/providers?features=streaming,vision&max_price=5&min_reliability=0.95
-
-# Response includes filtered results
-{
-    "providers": [...],
-    "filters_applied": {
-        "features": ["streaming", "vision"],
-        "max_price_sats_per_1k": 5,
-        "min_reliability": 0.95
-    },
-    "total_providers": 50,
-    "matching_providers": 12
-}
-```
-
-## Monitoring
-
-### Discovery Metrics
-
-Track discovery performance:
-
-```python
-class DiscoveryMetrics:
-    def __init__(self):
-        self.relay_health = {}
-        self.provider_count = 0
-        self.query_latency = []
-    
-    async def check_relay_health(self, relay_url: str):
-        """Monitor relay connectivity."""
-        start = time.time()
-        try:
-            await connect_to_relay(relay_url)
-            latency = time.time() - start
-            self.relay_health[relay_url] = {
-                'status': 'healthy',
-                'latency_ms': latency * 1000
-            }
-        except Exception as e:
-            self.relay_health[relay_url] = {
-                'status': 'unhealthy',
-                'error': str(e)
-            }
-```
-
-### Provider Monitoring
-
-```python
-@periodic_task(minutes=5)
-async def monitor_providers():
-    """Check provider health."""
-    providers = await discover_providers()
-    
-    for provider in providers:
-        try:
-            # Test endpoint
-            response = await test_provider_endpoint(provider.http_url)
-            
-            # Update metrics
-            await update_provider_metrics(
-                provider.npub,
-                success=response.status_code == 200,
-                response_time=response.elapsed
-            )
-        except Exception as e:
-            logger.warning(f"Provider {provider.name} check failed: {e}")
-```
-
-## Troubleshooting
-
-### No Providers Found
-
-```python
-# Debug discovery issues
-async def debug_discovery():
-    """Diagnose discovery problems."""
-    issues = []
-    
-    # Check relay connectivity
-    for relay in RELAYS:
-        if not await can_connect_to_relay(relay):
-            issues.append(f"Cannot connect to {relay}")
-    
-    # Check event publishing
-    if not await verify_own_events_visible():
-        issues.append("Own events not visible on relays")
-    
-    # Check filters
-    if len(await get_all_provider_events()) == 0:
-        issues.append("No provider events on any relay")
-    
-    return issues
-```
-
-### Relay Connection Issues
-
-```bash
-# Test relay connection
-wscat -c wss://relay.damus.io
-
-# Send subscription
-["REQ","test",{"kinds":[31990],"#d":["routstr"]}]
-```
-
-## Best Practices
-
-### For Providers
-
-1. **Consistent Identity**
-   - Use same npub across services
-   - Maintain profile metadata
-   - Verify identity on multiple platforms
-
-2. **Regular Updates**
-   - Publish status every few hours
-   - Update pricing promptly
-   - Remove stale information
-
-3. **Relay Diversity**
-   - Publish to 5+ relays
-   - Include regional relays
-   - Monitor relay health
-
-### For Clients
-
-1. **Verify Providers**
-   - Check multiple relays
-   - Verify endpoints work
-   - Monitor reliability over time
-
-2. **Cache Discovery**
-   - Cache provider list
-   - Refresh periodically
-   - Handle stale data gracefully
-
-3. **Fallback Options**
-   - Keep backup providers
-   - Handle discovery failures
-   - Support manual configuration
-
-## Next Steps
-
-- [Tor Support](tor.md) - Anonymous provider access
-- [Custom Pricing](custom-pricing.md) - Dynamic pricing strategies
-- [API Reference](../api/endpoints.md) - Discovery API details

docs/advanced/tor.md

@@ -1,530 +0,0 @@
-# Tor Support
-
-Routstr Core includes built-in support for Tor hidden services, enabling anonymous access to your API and enhanced privacy for users.
-
-## Overview
-
-Tor support provides:
-
-- **Anonymous Access**: Hidden service (.onion) address
-- **Enhanced Privacy**: No IP address logging
-- **Censorship Resistance**: Accessible from restricted networks
-- **Optional Usage**: Regular HTTP/HTTPS access remains available
-
-## Docker Setup
-
-### Using Docker Compose
-
-The included `compose.yml` automatically sets up Tor:
-
-```yaml
-version: '3.8'
-
-services:
-  routstr:
-    build: .
-    environment:
-      - TOR_PROXY_URL=socks5://tor:9050
-    ports:
-      - 8000:8000
-
-  tor:
-    image: ghcr.io/hundehausen/tor-hidden-service:latest
-    volumes:
-      - tor-data:/var/lib/tor
-    environment:
-      - HS_ROUTER=routstr:8000:80
-    depends_on:
-      - routstr
-
-volumes:
-  tor-data:
-```
-
-Start with:
-
-```bash
-docker compose up -d
-```
-
-### Getting Your Onion Address
-
-After starting, retrieve your hidden service address:
-
-```bash
-# View Tor logs
-docker compose logs tor
-
-# Or directly from the hostname file
-docker exec tor cat /var/lib/tor/hidden_service/hostname
-```
-
-Your onion address will look like:
-
-```
-roustrjfsdgfiueghsklchg.onion
-```
-
-## Manual Tor Setup
-
-### Install Tor
-
-```bash
-# Ubuntu/Debian
-sudo apt-get install tor
-
-# macOS
-brew install tor
-
-# Start Tor
-sudo systemctl start tor
-```
-
-### Configure Hidden Service
-
-Edit `/etc/tor/torrc`:
-
-```bash
-# Hidden service configuration
-HiddenServiceDir /var/lib/tor/routstr/
-HiddenServicePort 80 127.0.0.1:8000
-
-# Optional: Restrict to v3 addresses
-HiddenServiceVersion 3
-```
-
-Restart Tor:
-
-```bash
-sudo systemctl restart tor
-```
-
-Get onion address:
-
-```bash
-sudo cat /var/lib/tor/routstr/hostname
-```
-
-## Client Configuration
-
-### Using Tor with Python
-
-```python
-import httpx
-from openai import OpenAI
-
-# Configure SOCKS proxy
-proxies = {
-    "http://": "socks5://127.0.0.1:9050",
-    "https://": "socks5://127.0.0.1:9050"
-}
-
-# Create client with Tor
-http_client = httpx.Client(proxies=proxies)
-
-client = OpenAI(
-    api_key="sk-...",
-    base_url="http://roustrjfsdgfiueghsklchg.onion/v1",
-    http_client=http_client
-)
-
-# Use normally
-response = client.chat.completions.create(
-    model="gpt-3.5-turbo",
-    messages=[{"role": "user", "content": "Hello via Tor!"}]
-)
-```
-
-### Using Tor with cURL
-
-```bash
-# Install torify
-sudo apt-get install torsocks
-
-# Make request through Tor
-torify curl http://roustrjfsdgfiueghsklchg.onion/v1/models
-
-# Or with explicit proxy
-curl --socks5 127.0.0.1:9050 http://roustrjfsdgfiueghsklchg.onion/v1/models
-```
-
-### JavaScript/Node.js
-
-```javascript
-import { SocksProxyAgent } from 'socks-proxy-agent';
-import OpenAI from 'openai';
-
-// Create SOCKS agent
-const agent = new SocksProxyAgent('socks5://127.0.0.1:9050');
-
-// Configure OpenAI client
-const openai = new OpenAI({
-    apiKey: 'sk-...',
-    baseURL: 'http://roustrjfsdgfiueghsklchg.onion/v1',
-    httpAgent: agent
-});
-
-// Use normally
-const response = await openai.chat.completions.create({
-    model: 'gpt-3.5-turbo',
-    messages: [{ role: 'user', content: 'Hello via Tor!' }]
-});
-```
-
-## Configuration
-
-### Environment Variables
-
-Configure Tor proxy for outgoing connections:
-
-```bash
-# .env
-TOR_PROXY_URL=socks5://tor:9050  # Docker
-# or
-TOR_PROXY_URL=socks5://127.0.0.1:9050  # Local
-```
-
-### Publishing Onion Address
-
-Make your onion address discoverable:
-
-```bash
-# .env
-ONION_URL=http://roustrjfsdgfiueghsklchg.onion
-```
-
-This will be included in:
-
-- `/v1/info` endpoint
-- Nostr announcements
-- Admin dashboard
-
-## Security Considerations
-
-### Hidden Service Security
-
-1. **Keep Private Key Secure**
-
-   ```bash
-   # Backup hidden service keys
-   sudo tar -czf tor-keys-backup.tar.gz /var/lib/tor/routstr/
-   
-   # Restore to maintain same address
-   sudo tar -xzf tor-keys-backup.tar.gz -C /
-   ```
-
-2. **Access Control**
-
-   ```bash
-   # Restrict to authenticated clients
-   HiddenServiceAuthorizeClient stealth client1,client2
-   ```
-
-3. **Rate Limiting**
-
-   ```bash
-   # In torrc
-   HiddenServiceMaxStreams 100
-   HiddenServiceMaxStreamsCloseCircuit 1
-   ```
-
-### Operational Security
-
-1. **Separate Tor Instance**
-
-   ```yaml
-   # Use dedicated Tor container
-   tor:
-     image: ghcr.io/hundehausen/tor-hidden-service:latest
-     restart: always
-     networks:
-       - tor_network
-   ```
-
-2. **Monitor Tor Health**
-
-   ```python
-   async def check_tor_connection():
-       """Verify Tor connectivity."""
-       try:
-           async with httpx.AsyncClient(
-               proxies={"all://": TOR_PROXY_URL}
-           ) as client:
-               response = await client.get(
-                   "https://check.torproject.org/api/ip"
-               )
-               data = response.json()
-               return data.get("IsTor", False)
-       except Exception:
-           return False
-   ```
-
-3. **Logging Considerations**
-
-   ```python
-   # Don't log .onion addresses with IPs
-   def sanitize_logs(message: str) -> str:
-       # Remove IP addresses when .onion is present
-       if ".onion" in message:
-           message = re.sub(r'\d+\.\d+\.\d+\.\d+', '[IP]', message)
-       return message
-   ```
-
-## Performance Optimization
-
-### Connection Pooling
-
-```python
-# Reuse Tor circuits
-class TorConnectionPool:
-    def __init__(self, proxy_url: str):
-        self.proxy_url = proxy_url
-        self._clients = []
-    
-    async def get_client(self) -> httpx.AsyncClient:
-        if not self._clients:
-            client = httpx.AsyncClient(
-                proxies={"all://": self.proxy_url},
-                timeout=httpx.Timeout(30.0),
-                limits=httpx.Limits(
-                    max_keepalive_connections=5,
-                    max_connections=10
-                )
-            )
-            self._clients.append(client)
-        return self._clients[0]
-```
-
-### Circuit Management
-
-```python
-# Rotate Tor circuits periodically
-async def rotate_tor_circuit():
-    """Signal Tor to create new circuit."""
-    async with httpx.AsyncClient() as client:
-        # Tor control port (requires configuration)
-        response = await client.post(
-            "http://localhost:9051",
-            data="AUTHENTICATE\r\nSIGNAL NEWNYM\r\n"
-        )
-```
-
-### Caching Strategies
-
-```python
-# Cache responses for Tor users
-@lru_cache(maxsize=1000)
-def get_cached_response(
-    endpoint: str,
-    params_hash: str
-) -> Optional[dict]:
-    """Cache frequently accessed data."""
-    # Longer cache for Tor users due to latency
-    return cache.get(f"tor:{endpoint}:{params_hash}")
-```
-
-## Monitoring
-
-### Tor Metrics
-
-Track Tor-specific metrics:
-
-```python
-class TorMetrics:
-    def __init__(self):
-        self.tor_requests = 0
-        self.tor_errors = 0
-        self.circuit_builds = 0
-        self.average_latency = 0
-    
-    async def record_request(
-        self,
-        duration: float,
-        success: bool
-    ):
-        self.tor_requests += 1
-        if not success:
-            self.tor_errors += 1
-        
-        # Update average latency
-        self.average_latency = (
-            (self.average_latency * (self.tor_requests - 1) + duration)
-            / self.tor_requests
-        )
-```
-
-### Health Checks
-
-```python
-@router.get("/health/tor")
-async def tor_health():
-    """Check Tor service health."""
-    checks = {
-        "tor_proxy": await check_tor_proxy(),
-        "hidden_service": await check_hidden_service(),
-        "circuit_established": await check_circuit()
-    }
-    
-    status = "healthy" if all(checks.values()) else "unhealthy"
-    
-    return {
-        "status": status,
-        "checks": checks,
-        "metrics": {
-            "tor_requests_total": metrics.tor_requests,
-            "tor_error_rate": metrics.tor_errors / max(metrics.tor_requests, 1),
-            "average_latency_ms": metrics.average_latency * 1000
-        }
-    }
-```
-
-## Troubleshooting
-
-### Common Issues
-
-**Hidden Service Not Accessible**
-
-```bash
-# Check Tor logs
-docker compose logs tor
-# or
-sudo journalctl -u tor
-
-# Verify service is running
-sudo systemctl status tor
-
-# Test locally
-curl --socks5 127.0.0.1:9050 http://your-onion.onion/v1/info
-```
-
-**Slow Connection**
-
-- Tor adds 3+ hops of latency
-- Use connection pooling
-- Implement aggressive caching
-- Consider increasing timeouts
-
-**Connection Errors**
-
-```python
-# Implement Tor-specific retry logic
-async def tor_retry(func, max_retries=5):
-    for attempt in range(max_retries):
-        try:
-            return await func()
-        except httpx.ProxyError:
-            if attempt < max_retries - 1:
-                # Exponential backoff for circuit building
-                await asyncio.sleep(2 ** attempt)
-            else:
-                raise
-```
-
-### Debugging
-
-Enable Tor debug logging:
-
-```bash
-# In torrc
-Log debug file /var/log/tor/debug.log
-
-# Monitor in real-time
-tail -f /var/log/tor/debug.log
-```
-
-## Best Practices
-
-### For Operators
-
-1. **Backup Hidden Service Keys**
-   - Store securely offline
-   - Enables service recovery
-   - Maintains same .onion address
-
-2. **Monitor Tor Health**
-   - Check circuit establishment
-   - Track request latency
-   - Alert on failures
-
-3. **Separate Concerns**
-   - Run Tor in separate container
-   - Isolate from main application
-   - Use internal networks
-
-### For Users
-
-1. **Verify Onion Addresses**
-   - Check against multiple sources
-   - Bookmark verified addresses
-   - Watch for phishing
-
-2. **Handle Higher Latency**
-   - Increase client timeouts
-   - Implement retries
-   - Use connection pooling
-
-3. **Enhance Privacy**
-   - Use Tor Browser for web access
-   - Avoid mixing Tor/clearnet
-   - Don't include identifying info
-
-## Advanced Configuration
-
-### Multi-Hop Onion Services
-
-For extra security, chain multiple Tor instances:
-
-```yaml
-# compose.yml
-services:
-  tor-entry:
-    image: tor:latest
-    command: tor -f /etc/tor/torrc.entry
-    
-  tor-middle:
-    image: tor:latest
-    command: tor -f /etc/tor/torrc.middle
-    
-  tor-exit:
-    image: tor:latest
-    command: tor -f /etc/tor/torrc.exit
-```
-
-### Onion Service Authentication
-
-Require client authorization:
-
-```bash
-# Generate client auth
-openssl rand -base64 32 > client_auth_key
-
-# In torrc
-HiddenServiceDir /var/lib/tor/routstr/
-HiddenServicePort 80 127.0.0.1:8000
-HiddenServiceAuthorizeClient stealth payments
-```
-
-### Load Balancing
-
-Distribute load across multiple instances:
-
-```nginx
-# Onion service nginx config
-upstream routstr_backends {
-    server routstr1:8000;
-    server routstr2:8000;
-    server routstr3:8000;
-}
-
-server {
-    listen 80;
-    location / {
-        proxy_pass http://routstr_backends;
-    }
-}
-```
-
-## Next Steps
-
-- [Nostr Discovery](nostr.md) - Announce your onion service
-- [Docker Setup](../getting-started/docker.md) - Container configuration

docs/api/authentication.md

@@ -425,4 +425,4 @@ All API key usage is logged:
 
 - [Endpoints](endpoints.md) - Complete endpoint reference
 - [Errors](errors.md) - Error handling guide
-- [Using the API](../user-guide/using-api.md) - Integration examples
+- [Using the API](../client/integration.md) - Integration examples

docs/api/endpoints.md

@@ -434,6 +434,42 @@ Authorization: Bearer sk-...
 }
 ```
 
+### Create Child Key
+
+Creates one or more child API keys that share the parent's balance. Each child key creation costs a fixed amount (configurable).
+
+```http
+POST /v1/balance/child-key
+Authorization: Bearer sk-...
+```
+
+**Request Body:**
+
+```json
+{
+  "count": 1
+}
+```
+
+**Parameters:**
+
+| Parameter | Type | Required | Default | Description |
+|-----------|------|----------|---------|-------------|
+| `count` | integer | Yes | - | Number of child keys to create (1-50) |
+
+**Response:**
+
+```json
+{
+  "api_keys": ["sk-abc...", "sk-def..."],
+  "count": 2,
+  "cost_msats": 2000,
+  "cost_sats": 2,
+  "parent_balance": 98000,
+  "parent_balance_sats": 98
+}
+```
+
 ## Provider Discovery
 
 ## Admin Settings
@@ -527,4 +563,4 @@ Rate limit information is included in response headers.
 
 - [Errors](errors.md) - Error handling reference
 - [Authentication](authentication.md) - Auth details
-- [Examples](../user-guide/using-api.md) - Code examples
+- [Integration Guide](../client/integration.md) - Code examples

docs/api/errors.md

@@ -589,4 +589,4 @@ class ErrorMetrics:
 
 - [Authentication](authentication.md) - Auth error details
 - [Endpoints](endpoints.md) - Endpoint-specific errors
-- [Examples](../user-guide/using-api.md) - Error handling examples
+- [Integration Guide](../client/integration.md) - Error handling examples

docs/api/overview.md

@@ -99,19 +99,19 @@ All errors follow a consistent format:
 
 Standard OpenAI-compatible endpoints:
 
-- **Chat Completions**: `/v1/chat/completions`
-- **Completions**: `/v1/completions` *(Coming soon)*
-- **Embeddings**: `/v1/embeddings` *(Coming soon)*
-- **Images**: `/v1/images/generations` *(Coming soon)*
-- **Audio**: `/v1/audio/transcriptions` *(Coming soon)*
 - **Models**: `/v1/models`
+- **Responses**: `/v1/responses`
+- **Chat Completions**: `/v1/chat/completions`
+- **Embeddings**: `/v1/embeddings`
+- **Completions**: `/v1/completions` *(planned)*
+- **Images**: `/v1/images/generations` *(planned)*
+- **Audio**: `/v1/audio/transcriptions` *(planned)*
 
 ### Payment Endpoints
 
 Routstr-specific payment management:
 
-- **Wallet**: `/v1/wallet/*`
-- **Balance**: `/v1/balance`
+- **Balance**: `/v1/balance/*`
 - **Node Info**: `/v1/info`
 
 ### Admin Endpoints
@@ -137,9 +137,25 @@ Protected administrative functions:
 
 | Header | Description |
 |--------|-------------|
-| `X-Routstr-Version` | API version override |
 | `X-Cashu` | eCash token for per-request payment |
-| `X-Max-Cost` | Maximum acceptable cost in sats |
+
+#### X-Cashu: Stateless Per-Request Payment
+
+Instead of using `Authorization: Bearer sk-...`, you can send a Cashu token directly in the `X-Cashu` header. The response will include an `X-Cashu-Refund` header with your change.
+
+```bash
+curl https://api.routstr.com/v1/chat/completions \
+  -H "X-Cashu: cashuA3s8jKx9..." \
+  -H "Content-Type: application/json" \
+  -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Hello"}]}'
+```
+
+The response includes your change in the same header:
+```
+X-Cashu: cashuA7k2mNp4...
+```
+
+This is fully stateless—no session, no `/v1/balance/refund` call needed. However, **streaming does not work with `X-Cashu`** because the refund can only be calculated after the full response is generated.
 
 ## Response Headers
 
@@ -149,16 +165,8 @@ Protected administrative functions:
 |--------|-------------|
 | `Content-Type` | Response format |
 | `Content-Length` | Response size |
-| `X-Routstr-Request-ID` | Unique request identifier |
-| `X-Routstr-Version` | API version used |
-
-### Cost Headers
-
-| Header | Description |
-|--------|-------------|
-| `X-Routstr-Cost` | Request cost in sats |
-| `X-Routstr-Balance` | Remaining balance |
-| `X-Cashu` | Change token (if applicable) |
+| `X-Request-ID` | Unique request identifier |
+| `X-Cashu` | Change token (when request used `X-Cashu` header) |
 
 ## Streaming Responses
 
@@ -245,8 +253,6 @@ X-Webhook-Signature: sha256=...
 
 - Current version: `v1`
 - Version in URL path: `/v1/endpoint`
-- Override with header: `X-Routstr-Version: v2`
-- Deprecation notices: 6 months
 
 ## Status Codes
 
@@ -284,82 +290,23 @@ Responses are compressed with gzip when:
 - Response is larger than 1KB
 - Content type is compressible
 
-## Pagination
+## Batch Requests *(planned)*
 
-List endpoints support pagination:
+Process multiple operations in one request. Coming soon.
+
+## Node Info
+
+Get node metadata:
 
 ```
-GET /v1/transactions?limit=50&offset=100
+GET /v1/info
 ```
 
-Response includes pagination metadata:
-
-```json
-{
-  "data": [...],
-  "has_more": true,
-  "total": 500,
-  "limit": 50,
-  "offset": 100
-}
-```
-
-## Field Filtering
-
-Select specific fields in responses:
-
-```
-GET /v1/models?fields=id,name,pricing
-```
-
-## Batch Requests
-
-Process multiple operations in one request:
-
-```json
-POST /v1/batch
-{
-  "requests": [
-    {"method": "POST", "endpoint": "/chat/completions", "body": {...}},
-    {"method": "GET", "endpoint": "/models"},
-    {"method": "GET", "endpoint": "/balance"}
-  ]
-}
-```
-
-## Idempotency
-
-Prevent duplicate operations:
-
-```
-Idempotency-Key: unique-request-id
-```
-
-Keys are stored for 24 hours.
-
-## Health Check
-
-Monitor service status:
-
-```
-GET /health
-
-Response:
-{
-  "status": "healthy",
-  "version": "0.2.0",
-  "timestamp": "2024-01-01T00:00:00Z",
-  "checks": {
-    "database": "ok",
-    "upstream": "ok",
-    "mint": "ok"
-  }
-}
-```
+Supported models and pricing are available at `/v1/models`.
 
 ## Next Steps
 
 - [Authentication](authentication.md) - Detailed auth guide
 - [Endpoints](endpoints.md) - Complete endpoint reference
 - [Errors](errors.md) - Error handling guide
-- [Examples](../user-guide/using-api.md) - Code examples
+- [Integration Guide](../client/integration.md) - Code examples

docs/client/integration.md

@@ -0,0 +1,143 @@
+# Using the API
+
+Routstr is **OpenAI-compatible**. Almost any AI application, SDK, or tool that supports custom endpoints will work out of the box. Just change two things:
+
+```
+BASE_URL  →  https://api.routstr.com/v1
+API_KEY   →  sk-... or cashuA...
+```
+
+**Both work as API keys:**
+- `sk-7f8e9d...` — Session key (from Lightning invoice or Cashu import)
+- `cashuA3s8j...` — Raw Cashu token (use directly from your wallet)
+
+If the app lets you set a base URL and API key, you're good to go.
+
+---
+
+## Quick Setup Examples
+
+### OpenAI SDK (Python/JS)
+
+```python
+client = OpenAI(base_url="https://api.routstr.com/v1", api_key="sk-...")  # or any provider's URL
+```
+
+### Claude Code
+
+```bash
+export ANTHROPIC_BASE_URL=https://api.routstr.com/v1
+export ANTHROPIC_AUTH_TOKEN=sk-...
+```
+
+### Any OpenAI-compatible app
+
+Look for "Custom API endpoint", "Base URL", or "OpenAI-compatible" in settings. Paste the URL and key.
+
+---
+
+## Detailed Examples
+
+### Python (Official SDK)
+
+```python
+from openai import OpenAI
+
+# 1. Initialize with Routstr URL and your funded key
+client = OpenAI(
+    base_url="https://api.routstr.com/v1",
+    api_key="sk-7f8e9d..." 
+)
+
+# 2. Call the API normally
+response = client.chat.completions.create(
+    model="gpt-4o",
+    messages=[{"role": "user", "content": "Hello!"}],
+    stream=True
+)
+
+for chunk in response:
+    if chunk.choices[0].delta.content:
+        print(chunk.choices[0].delta.content, end="")
+```
+
+### Node.js
+
+```javascript
+import OpenAI from 'openai';
+
+// You can use a session key OR a raw Cashu token directly
+const openai = new OpenAI({
+  baseURL: 'https://api.routstr.com/v1',
+  apiKey: 'cashuA3s8jKx9...', // or 'sk-7f8e9d...'
+});
+
+async function main() {
+  const completion = await openai.chat.completions.create({
+    messages: [{ role: 'user', content: 'Say this is a test' }],
+    model: 'gpt-3.5-turbo',
+  });
+
+  console.log(completion.choices[0]);
+}
+
+main();
+```
+
+### cURL
+
+```bash
+# Works with session key or raw Cashu token
+curl https://api.routstr.com/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer cashuA3s8jKx9..." \
+  -d '{
+    "model": "gpt-4o-mini",
+    "messages": [{"role": "user", "content": "Hello!"}]
+  }'
+```
+
+---
+
+## Error Handling
+
+### Insufficient Balance (402 Payment Required)
+If your session runs out of funds, the API will return a `402` error.
+
+```json
+{
+  "error": {
+    "message": "Insufficient balance. Current: 1000 msat, Required: 5000 msat",
+    "type": "insufficient_balance",
+    "code": 402
+  }
+}
+```
+
+**Action**: Top up your key using the `/lightning/invoice` (topup purpose) or `/v1/balance/topup` endpoints.
+
+### Rate Limiting
+Routstr passes through rate limits from the upstream provider. Handle `429 Too Many Requests` with standard exponential backoff.
+
+---
+
+## Advanced: Tor Access
+
+If the node is running as a hidden service, use a SOCKS5 proxy (like `127.0.0.1:9050`).
+
+**Python:**
+```python
+import httpx
+from openai import OpenAI
+
+proxy_mounts = {
+    "http://": httpx.HTTPTransport(proxy="socks5://127.0.0.1:9050"),
+    "https://": httpx.HTTPTransport(proxy="socks5://127.0.0.1:9050"),
+}
+
+client = OpenAI(
+    base_url="http://verylongonionaddress.onion/v1",
+    api_key="sk-...",
+    http_client=httpx.Client(mounts=proxy_mounts),
+)
+```

docs/client/introduction.md

@@ -0,0 +1,142 @@
+# Introduction to Routstr
+
+Welcome to the Routstr Core User Guide. This guide will help you understand how to use Routstr to access AI APIs with Bitcoin micropayments.
+
+## What You'll Learn
+
+- How the payment system works (Cashu eCash)
+- Creating and managing API keys (Ephemeral Sessions)
+- Making API calls through Routstr
+- Using the admin dashboard
+
+## Prerequisites
+
+### 💰 Wallet
+
+Cashu ([cashu.me](https://cashu.me)) or Lightning ([Strike](https://strike.me), Cash App, etc.)
+
+### 🌐 Provider
+
+A Routstr node, e.g. `https://api.routstr.com`
+
+### 🤖 Client
+
+OpenAI SDK, Claude Code, Cursor, or any OpenAI-compatible tool
+
+---
+
+## How Routstr Works
+
+Routstr is a **Payment Proxy**. It sits between your code and the AI provider.
+
+### Traditional API vs Routstr
+
+| Traditional | Routstr |
+|---|---|
+| Credit Card Required | Bitcoin / Lightning / eCash |
+| Monthly Billing | Pay-per-request (Real-time) |
+| KYC / Account | No Account / Private |
+| Single Provider | Aggregated Providers |
+
+### Key Concepts
+
+#### 1. Cashu eCash
+
+Digital bearer tokens backed by Bitcoin. They are instant, private, and have no fees for internal transfers. Routstr uses these tokens as the "credits" for API requests.
+
+#### 2. Ephemeral Sessions (API Keys)
+
+Instead of a permanent account, you create a **Session**.
+
+- You fund a session with eCash or Lightning.
+- Routstr gives you an `api_key` (`sk-...`) representing that session.
+- You use the `api_key` until funds run out or you finish your task.
+- You can **refund** the remaining balance back to your wallet at any time.
+
+#### 3. Millisats (msats)
+
+Everything is priced in **millisatoshis**.
+
+- 1 Satoshi (sat) = 1,000 msats.
+- This allows for extremely precise pricing (e.g., 0.05 sats per prompt).
+
+---
+
+## Workflow: Zero to Intelligence
+
+### 1. Fund a Session
+
+You need an `api_key` with a balance.
+
+**Easiest: Use the Web UI**
+Visit the node's root page (e.g., [api.routstr.com](https://api.routstr.com)) or [chat.routstr.com](https://chat.routstr.com) → Settings to create a key visually with Lightning.
+
+**Option A: Lightning Invoice (CLI)**
+Generate an invoice and pay it with any Lightning wallet.
+
+```bash
+curl -X POST https://api.routstr.com/lightning/invoice \
+  -d '{"amount_sats": 1000, "purpose": "create"}'
+```
+
+*Returns an invoice (`bolt11`) and an ID. Once paid, the status endpoint returns your `api_key`.*
+
+**Option B: Cashu Token (Best for privacy & devs)**
+If you have a Cashu wallet, you can copy a token string (`cashuA...`) and use it directly.
+
+- **Direct Usage**: Use the token *as* your API key in the `Authorization` header.
+- **Import**: Or exchange it for a standard `sk-...` key:
+
+```bash
+curl "https://api.routstr.com/v1/balance/create?initial_balance_token=cashuA..."
+```
+
+*Returns your `api_key` immediately.*
+
+### 2. Configure Your Client
+
+Use the standard OpenAI SDK, just changing the `base_url` and `api_key`.
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="https://api.routstr.com/v1",
+    api_key="sk-7f8e9d..."  # The key from Step 1
+)
+```
+
+### 3. Make Requests
+
+```python
+response = client.chat.completions.create(
+    model="gpt-4o",
+    messages=[{"role": "user", "content": "Explain quantum computing."}]
+)
+```
+
+### 4. Withdraw Change
+
+When you are done, get your change back as a Cashu token.
+
+```bash
+curl -X POST https://api.routstr.com/v1/balance/refund \
+  -H "Authorization: Bearer sk-7f8e9d..."
+```
+
+*Returns a `token` that you can paste back into Nutstash or Minibits to reclaim your funds.*
+
+---
+
+## Supported Features
+
+- **Responses**: `/v1/responses` (OpenAI Responses API)
+- **Chat Completions**: `/v1/chat/completions` (Streaming supported)
+- **Embeddings**: `/v1/embeddings`
+- **Models**: `/v1/models` (List available models and prices)
+
+## Next Steps
+
+- **[Payment Flow](payments.md)**: Detailed breakdown of the funding lifecycle.
+- **[Models & Pricing](../provider/pricing.md)**: How costs are calculated.
+- **[Admin Dashboard](../provider/dashboard.md)**: Managing your node if you are the operator.

docs/client/payments.md

@@ -0,0 +1,97 @@
+# Payment Flow
+
+Routstr uses a **Pre-paid, Ephemeral** payment model. Pay first, use the funds, withdraw the rest. No accounts, no credit cards, no trails.
+
+```
+💰 Deposit  →  🤖 Use AI  →  💸 Withdraw Change
+```
+
+## 1. Creating a Balance (Deposit)
+
+To start making requests, you must create a "Balance" (represented by an API Key).
+
+### Method A: Lightning Network (Bolt11)
+
+**Ideal for**: Users connecting from a standard Lightning wallet (Strike, Cash App, WoS).
+
+1. **Request Invoice**:
+    `POST /lightning/invoice` with `{"amount_sats": 5000, "purpose": "create"}`.
+2. **Pay Invoice**: User scans and pays the QR code/bolt11 string.
+3. **Receive Key**: Routstr detects the payment and issues a new API Key (`sk-...`) pre-loaded with 5,000 sats (5,000,000 msats).
+
+### Method B: Cashu Token Import
+
+**Ideal for**: Private, instant access or automated agents.
+
+1. **Generate Token**: User creates a token in their local wallet (e.g., 1000 sats).
+2. **Import**: `GET /v1/balance/create?initial_balance_token=cashuA...`
+3. **Receive Key**: Routstr claims the token and issues an API Key (`sk-...`) with that balance.
+
+---
+
+## 2. Consuming Funds (Inference)
+
+Every time you make a request to `/v1/chat/completions` (or others), the cost is deducted from your balance **in real-time**.
+
+### Cost Calculation
+
+`Cost = (Input_Tokens * Price_Input) + (Output_Tokens * Price_Output) + Request_Fee`
+
+- Prices are defined per model (see `/v1/models`).
+- If you stream the response, the balance is deducted incrementally or finalized at the end of the stream.
+- If your balance hits 0 mid-stream, the connection is closed.
+
+### Headers
+
+Routstr checks the `Authorization: Bearer sk-...` header to identify which balance to charge.
+
+---
+
+## 3. Topping Up
+
+If your balance runs low, you don't need a new key. You can top up the existing one.
+
+### Via Lightning
+
+`POST /lightning/invoice` with `{"amount_sats": 1000, "purpose": "topup", "api_key": "sk-..."}`.
+*Once paid, the funds are added to your existing key.*
+
+### Via Cashu
+
+`POST /v1/balance/topup` with `{"cashu_token": "..."}` and `Authorization: Bearer sk-...`.
+
+---
+
+## 4. Refund (Withdrawal)
+
+Don't leave large balances sitting on a node—it's a hot wallet. When you're done, get your sats back.
+
+### Endpoint
+
+`POST /v1/balance/refund`
+
+**Headers**:
+`Authorization: Bearer sk-...`
+
+**Response**:
+
+```json
+{
+  "token": "cashuAeyJ0b2tlbiI6W3sibWludCI6...",
+  "msats": "450000"
+}
+```
+
+You can verify the refund was successful by checking that the API Key is now invalid or has 0 balance. Copy the `token` string and paste it into your Cashu wallet to claim the Bitcoin.
+
+---
+
+## Summary
+
+| Step | Action | Result |
+|------|--------|--------|
+| 💰 **Deposit** | Pay Lightning invoice or import Cashu | Get `sk-...` key |
+| 🤖 **Use** | Make API requests | Balance decreases |
+| 💸 **Refund** | Call `/v1/balance/refund` | Get Cashu token back |
+
+That's it. No monthly bills, no surprise charges, no data harvesting.

docs/contributing/architecture.md

@@ -4,7 +4,7 @@ This document describes the high-level architecture of Routstr Core, helping con
 
 ## System Overview
 
-Routstr Core is a FastAPI-based reverse proxy that adds Bitcoin micropayments to OpenAI-compatible APIs.
+Routstr Core is a FastAPI-based reverse proxy that adds Bitcoin micropayments to OpenAI-compatible APIs and can optionally announce providers via Nostr.
 
 ```mermaid
 graph TB
@@ -20,7 +20,7 @@ graph TB
         Auth[Auth Module]
         Payment[Payment Module]
         Proxy[Proxy Module]
-        DB[(SQLite DB)]
+        DB[(SQLModel DB)]
         
         API --> Auth
         Auth --> Payment
@@ -40,57 +40,87 @@ graph TB
 
 The main application is initialized in `routstr/core/main.py`:
 
-- **Lifespan Management**: Handles startup/shutdown tasks
-- **Middleware**: CORS, logging, error handling
-- **Routers**: Modular endpoint organization
-- **Background Tasks**: Price updates, automatic payouts
+- **Lifespan Management**: Runs migrations, initializes DB, refreshes pricing/models, starts background tasks
+- **Middleware**: CORS and request logging
+- **Routers**: Admin, pricing/models, balance/wallet, providers discovery, proxy
+- **Background Tasks**: Price refresh, model map refresh, payouts, node announcements, provider discovery refresh
 
 ### Authentication System
 
 Located in `routstr/auth.py`, handles:
 
-- **API Key Validation**: Hashed key storage and lookup
-- **Balance Checking**: Ensures sufficient funds
-- **Rate Limiting**: Optional request throttling
-- **Token Redemption**: Converts eCash to balance
+- **API Key Validation**: SHA-256 hashed key lookup and persistence
+- **Balance Checking**: Ensures sufficient funds before requests
+- **Token Redemption**: Converts Cashu tokens to balance
 
 ### Payment Processing
 
 The `routstr/payment/` module manages:
 
 - **Cost Calculation**: Token-based or fixed pricing
-- **Model Pricing**: Dynamic pricing from models.json
-- **Currency Conversion**: BTC/USD rate management
-- **Fee Application**: Exchange and provider fees
+- **Model Pricing**: Derived from upstream providers and DB overrides
+- **Currency Conversion**: BTC/USD price refresh and conversion
+- **Fee Application**: Provider fee applied to upstream model pricing
 
 ### Request Proxying
 
 `routstr/proxy.py` handles:
 
-- **Request Forwarding**: Preserves headers and body
-- **Response Streaming**: Efficient memory usage
-- **Usage Tracking**: Counts tokens and costs
-- **Error Handling**: Graceful upstream failures
+- **Request Forwarding**: Forwards requests to selected upstream providers
+- **Response Streaming**: Streaming and non-streaming paths
+- **Usage Tracking**: Adjusts costs after upstream responses
+- **Error Handling**: Maps upstream errors to consistent responses
 
 ### Database Layer
 
 Using SQLModel in `routstr/core/db.py`:
 
 ```python
-# Core models
-APIKey:
-  - id: Primary key
-  - key_hash: Hashed API key
+# Core tables
+ApiKey:
+  - hashed_key: Primary key (SHA-256 of key or Cashu token)
   - balance: Current balance (msats)
-  - created_at: Timestamp
-  - metadata: JSON field
+  - reserved_balance: Reserved balance (msats)
+  - refund_address: Optional LNURL for refunds
+  - key_expiry_time: Optional refund expiry timestamp
+  - total_spent: Total spent (msats)
+  - total_requests: Request count
+  - refund_mint_url: Mint URL for refunds
+  - refund_currency: Refund currency
 
-Transaction:
+UpstreamProviderRow:
   - id: Primary key
-  - api_key_id: Foreign key
-  - amount: Transaction amount
-  - type: deposit/usage/withdrawal
-  - timestamp: When occurred
+  - provider_type: openai/anthropic/azure/openrouter/etc.
+  - base_url: Provider API base URL
+  - api_key: Provider API key
+  - api_version: Optional API version
+  - enabled: Provider enabled flag
+  - provider_fee: Provider fee multiplier
+
+ModelRow:
+  - id: Model ID
+  - upstream_provider_id: Provider foreign key
+  - name: Model name
+  - architecture: JSON
+  - pricing: JSON
+  - sats_pricing: JSON
+  - per_request_limits: JSON
+  - top_provider: JSON
+  - canonical_slug: Canonical model slug
+  - alias_ids: Model aliases
+  - enabled: Model enabled flag
+
+LightningInvoice:
+  - id: Primary key
+  - bolt11: Invoice
+  - amount_sats: Amount in sats
+  - payment_hash: Payment hash
+  - status: pending/paid/expired/cancelled
+  - api_key_hash: Optional associated API key
+  - purpose: create/topup
+  - created_at: Unix timestamp
+  - expires_at: Unix timestamp
+  - paid_at: Unix timestamp
 ```
 
 ## Request Flow
@@ -107,10 +137,10 @@ sequenceDiagram
     C->>R: API Request + Key
     R->>D: Validate Key
     D-->>R: Key Info + Balance
-    R->>R: Check Balance
+    R->>R: Reserve Max Cost
     R->>P: Forward Request
     P-->>R: AI Response
-    R->>D: Deduct Cost
+    R->>D: Finalize Cost (adjust by usage)
     R-->>C: Return Response
 ```
 
@@ -124,36 +154,20 @@ sequenceDiagram
     participant M as Cashu Mint
     participant D as Database
     
-    C->>R: Create Key Request + Token
-    R->>W: Validate Token
+    C->>R: Request + Cashu Token
+    R->>W: Redeem Token
     W->>M: Verify with Mint
     M-->>W: Token Valid
     W-->>R: Token Amount
-    R->>D: Create Key + Balance
-    R-->>C: Return API Key
+    R->>D: Create/Update Key + Balance
+    R-->>C: Continue Request
 ```
 
 ## Key Design Decisions
 
 ### 1. Async Architecture
 
-Everything is async for maximum performance:
-
-```python
-async def handle_request(request: Request) -> Response:
-    # Non-blocking database queries
-    api_key = await get_api_key(request.headers["Authorization"])
-    
-    # Concurrent operations
-    balance_check, rate_limit = await asyncio.gather(
-        check_balance(api_key),
-        check_rate_limit(api_key)
-    )
-    
-    # Stream response without blocking
-    async for chunk in proxy_request(request):
-        yield chunk
-```
+The system is async end-to-end, with background tasks for pricing refresh, provider discovery, model map refresh, and payouts.
 
 ### 2. Modular Design
 
@@ -166,82 +180,46 @@ Components are loosely coupled:
 
 ### 3. Error Handling
 
-Graceful degradation and clear error messages:
-
-```python
-class RoustrError(Exception):
-    """Base exception with structured error response"""
-    status_code: int = 500
-    error_type: str = "internal_error"
-    
-class InsufficientBalanceError(RoustrError):
-    status_code = 402
-    error_type = "insufficient_balance"
-```
+Exceptions are handled by FastAPI exception handlers to return consistent JSON responses with a request ID.
 
 ### 4. Database Migrations
 
-Using Alembic for schema management:
-
-- Auto-migrations on startup
-- Version control for schema changes
-- Rollback capability
-- Zero-downtime updates
+Alembic migrations are run on startup, and tables are created for any models not tracked by migrations.
 
 ## Security Architecture
 
 ### API Key Security
 
 - **Storage**: SHA-256 hashed keys
-- **Generation**: Cryptographically secure random
-- **Validation**: Constant-time comparison
-- **Rotation**: Support for key expiry
+- **Generation**: Cryptographically secure random (when creating new keys)
+- **Validation**: Hash lookup in DB
+- **Expiry**: Optional refund flow via `key_expiry_time` and `refund_address`
 
 ### Payment Security
 
-- **Token Validation**: Cryptographic verification
-- **Double-Spend Prevention**: Mint verification
-- **Balance Protection**: Atomic transactions
-- **Audit Trail**: All transactions logged
+- **Token Validation**: Cashu token redemption via mint
+- **Balance Protection**: Atomic updates and reserved balance tracking
+- **Audit Trail**: Structured logging of payments and adjustments
 
 ### Network Security
 
-- **HTTPS**: Enforced in production
 - **CORS**: Configurable origins
-- **Rate Limiting**: Per-key limits
 - **Input Validation**: Pydantic models
 
 ## Performance Considerations
 
 ### Caching Strategy
 
-```python
-# Model pricing cache
-@lru_cache(maxsize=100)
-def get_model_price(model_id: str) -> ModelPrice:
-    return MODELS.get(model_id)
-
-# Balance cache with TTL
-balance_cache = TTLCache(maxsize=1000, ttl=60)
-```
+Model and provider selections are cached in process memory and refreshed on a schedule.
 
 ### Database Optimization
 
-- **Connection Pooling**: Reuse connections
-- **Indexed Queries**: Key lookups are O(1)
-- **Batch Operations**: Group updates
 - **Async I/O**: Non-blocking queries
+- **Atomic Updates**: Balance reservation and finalization updates
 
 ### Streaming Responses
 
-Efficient memory usage for large responses:
-
-```python
-async def stream_response(upstream_response):
-    async for chunk in upstream_response.aiter_bytes():
-        # Process chunk without loading full response
-        yield process_chunk(chunk)
-```
+Streaming responses are forwarded from upstream providers with usage tracking hooks.
 
 ## Extension Points
 
@@ -273,8 +251,6 @@ async def stream_response(upstream_response):
 
 - Mock external dependencies
 - Test business logic in isolation
-- Fast execution (< 1 second per test)
-- High coverage target (> 80%)
 
 ### Integration Tests
 
@@ -285,10 +261,7 @@ async def stream_response(upstream_response):
 
 ### Performance Tests
 
-- Load testing with locust
-- Memory profiling
-- Database query optimization
-- Response time benchmarks
+- Response time benchmarks (as needed)
 
 ## Monitoring and Observability
 
@@ -308,41 +281,17 @@ logger.info("api_request", extra={
 
 ### Metrics Collection
 
-Key metrics tracked:
-
-- Request rate by endpoint
-- Token usage by model
-- Balance changes
-- Error rates
-- Response times
+Structured logs are emitted for requests, pricing, and payment events.
 
 ### Health Checks
 
-```python
-@app.get("/health")
-async def health_check():
-    return {
-        "status": "healthy",
-        "version": __version__,
-        "database": await check_db(),
-        "upstream": await check_upstream(),
-        "mints": await check_mints()
-    }
-```
+Use `/v1/info` for basic service metadata and configuration visibility.
 
 ## Deployment Architecture
 
 ### Container Structure
 
-```dockerfile
-# Multi-stage build
-FROM python:3.11-slim AS builder
-# Install dependencies
-
-FROM python:3.11-slim
-# Copy only runtime needs
-# Run as non-root user
-```
+See `core/Dockerfile` for the current container build configuration.
 
 ### Environment Configuration
 
@@ -352,10 +301,9 @@ FROM python:3.11-slim
 
 ### Scaling Considerations
 
-- **Horizontal**: Multiple instances behind load balancer
+- **Horizontal**: Multiple instances behind a load balancer
 - **Vertical**: Async handles high concurrency
-- **Database**: Consider PostgreSQL for scale
-- **Caching**: Redis for distributed cache
+- **Database**: Configure `DATABASE_URL` for external databases
 
 ## Future Architecture
 

docs/contributing/code-structure.md

@@ -7,10 +7,13 @@ This guide provides a detailed overview of Routstr Core's codebase organization
 ```
 routstr-core/
 ├── routstr/                    # Main application package
-│   ├── __init__.py            # Package initialization, loads .env
-│   ├── auth.py                # Authentication and authorization
+│   ├── __init__.py            # Package initialization, exports FastAPI app
+│   ├── algorithm.py           # Model selection/mapping logic
+│   ├── auth.py                # Bearer/Cashu auth and payment handling
 │   ├── balance.py             # Balance management endpoints
 │   ├── discovery.py           # Nostr relay discovery
+│   ├── lightning.py           # Lightning invoice topups
+│   ├── nip91.py               # Node announcement logic
 │   ├── proxy.py               # Request proxying logic
 │   ├── wallet.py              # Cashu wallet operations
 │   │
@@ -18,19 +21,23 @@ routstr-core/
 │   │   ├── __init__.py
 │   │   ├── admin.py          # Admin dashboard and API
 │   │   ├── db.py             # Database models and connection
-│   │   ├── exceptions.py     # Custom exception classes
+│   │   ├── exceptions.py     # Exception handlers
 │   │   ├── logging.py        # Structured logging setup
 │   │   ├── main.py           # FastAPI app initialization
 │   │   └── middleware.py     # HTTP middleware components
 │   │
-│   └── payment/               # Payment processing
-│       ├── __init__.py
-│       ├── cost_calculation.py # Usage cost calculation
-│       ├── helpers.py         # Payment utilities
-│       ├── lnurl.py          # Lightning URL support
-│       ├── models.py         # Model pricing management
-│       ├── price.py          # BTC/USD price handling
-│       └── x_cashu.py        # Cashu header protocol
+│   ├── payment/               # Payment processing
+│   │   ├── __init__.py
+│   │   ├── cost_calculation.py # Usage cost calculation
+│   │   ├── helpers.py         # Payment utilities
+│   │   ├── lnurl.py          # Lightning URL support
+│   │   ├── models.py         # Model pricing management
+│   │   └── price.py          # BTC/USD price handling
+│   │
+│   └── upstream/              # Upstream provider integrations
+│       ├── base.py           # Base provider logic
+│       ├── helpers.py        # Provider init and model refresh
+│       └── ...               # Provider implementations
 │
 ├── tests/                     # Test suite
 │   ├── __init__.py
@@ -45,7 +52,12 @@ routstr-core/
 │   └── versions/             # Migration files
 │
 ├── scripts/                  # Utility scripts
-│   └── models_meta.py       # Fetch model pricing
+│   ├── models_meta.py       # Fetch model pricing
+│   └── ...                  # Build/update helpers
+│
+├── examples/                 # Example clients
+├── testing-clients/          # HTML test clients
+├── ui/                       # Next.js admin UI
 │
 ├── docs/                    # Documentation
 ├── logs/                    # Application logs (git ignored)
@@ -71,30 +83,27 @@ routstr-core/
 #### `routstr/__init__.py`
 
 ```python
-# Loads environment variables
-import dotenv
-dotenv.load_dotenv()
-
-# Exports FastAPI app
 from .core.main import app as fastapi_app
+
+__all__ = ["fastapi_app"]
 ```
 
 #### `routstr/core/main.py`
 
 ```python
 # FastAPI application setup
-app = FastAPI(
-    title="Routstr Node",
-    lifespan=lifespan,  # Manages startup/shutdown
-)
+app = FastAPI(version=__version__, lifespan=lifespan)
 
 # Middleware registration
 app.add_middleware(CORSMiddleware, ...)
 app.add_middleware(LoggingMiddleware)
 
 # Router inclusion
+app.include_router(models_router)
 app.include_router(admin_router)
 app.include_router(balance_router)
+app.include_router(deprecated_wallet_router)
+app.include_router(providers_router)
 app.include_router(proxy_router)
 ```
 
@@ -102,29 +111,24 @@ app.include_router(proxy_router)
 
 #### `routstr/auth.py`
 
-Handles API key validation and authorization:
+Handles bearer key validation and payment lifecycle (bearer or Cashu token):
 
 ```python
-class APIKeyAuth:
-    """FastAPI dependency for API key authentication"""
-    
-    async def __call__(self, request: Request) -> APIKey:
-        # Extract and validate API key
-        # Check balance
-        # Return authenticated key object
-
-# Usage in routes:
-@router.get("/protected")
-async def protected_route(api_key: APIKey = Depends(APIKeyAuth())):
-    pass
+async def validate_bearer_key(
+    bearer_key: str,
+    session: AsyncSession,
+    refund_address: Optional[str] = None,
+    key_expiry_time: Optional[int] = None,
+) -> ApiKey:
+    """Validate bearer API key or redeem Cashu token into a balance."""
 ```
 
 Key functions:
 
-- `create_api_key()` - Generate new API keys
-- `validate_api_key()` - Verify and retrieve key
-- `check_balance()` - Ensure sufficient funds
-- `update_last_used()` - Track usage
+- `validate_bearer_key()` - Validate API key or Cashu token
+- `pay_for_request()` - Reserve max cost before upstream call
+- `adjust_payment_for_tokens()` - Adjust final cost after response
+- `revert_pay_for_request()` - Refund on upstream failure
 
 ### Payment Processing
 
@@ -133,56 +137,30 @@ Key functions:
 Calculates request costs:
 
 ```python
-def calculate_request_cost(
-    model: str,
-    prompt_tokens: int,
-    completion_tokens: int,
-    **kwargs
-) -> CostData:
-    """Calculate cost in millisatoshis"""
-    # Model-based or fixed pricing
-    # Token counting
-    # Fee application
-    # Currency conversion
+async def calculate_cost(
+    response_data: dict, max_cost: int, session: AsyncSession
+) -> CostData | MaxCostData | CostDataError:
+    """Calculate cost in millisatoshis from response usage or model pricing."""
 ```
 
 #### `routstr/payment/models.py`
 
-Manages model pricing data:
+Manages model pricing, database overrides, and pricing refresh:
 
 ```python
-class ModelPrice:
+class Model(BaseModel):
     id: str
     name: str
-    pricing: dict[str, float]  # USD prices
-    context_length: int
-    
-# Global model registry
-MODELS: dict[str, ModelPrice] = load_models()
+    pricing: Pricing
+    sats_pricing: Pricing | None = None
 
-# Dynamic price updates
 async def update_sats_pricing():
-    """Background task to update BTC prices"""
+    """Periodic task to update sats pricing for providers and overrides."""
 ```
 
-#### `routstr/payment/x_cashu.py`
+#### `routstr/proxy.py` + `routstr/upstream/*`
 
-Implements Cashu payment protocol:
-
-```python
-class XCashuHandler:
-    """Handle x-cashu header payments"""
-    
-    async def process_request_payment(
-        self, 
-        token: str,
-        estimated_cost: int
-    ) -> PaymentResult:
-        # Validate token
-        # Check minimum amount
-        # Process payment
-        # Generate change
-```
+The `x-cashu` header is handled by the proxy route and delegated to upstream providers.
 
 ### Request Proxying
 
@@ -191,17 +169,11 @@ class XCashuHandler:
 Core proxy functionality:
 
 ```python
-@router.api_route("/{path:path}", methods=ALL_METHODS)
-async def proxy_request(
-    request: Request,
-    path: str,
-    api_key: APIKey = Depends(APIKeyAuth())
-) -> Response:
-    """Forward requests to upstream provider"""
-    # Build upstream request
-    # Stream response
-    # Track usage
-    # Deduct costs
+@proxy_router.api_route("/{path:path}", methods=["GET", "POST"], response_model=None)
+async def proxy(
+    request: Request, path: str, session: AsyncSession = Depends(get_session)
+) -> Response | StreamingResponse:
+    """Forward requests to upstream provider and charge usage."""
 ```
 
 Key features:
@@ -215,27 +187,29 @@ Key features:
 
 #### `routstr/core/db.py`
 
-SQLModel definitions:
+SQLModel definitions (selected):
 
 ```python
-class APIKey(SQLModel, table=True):
-    id: int | None = Field(primary_key=True)
-    key_hash: str = Field(index=True, unique=True)
-    balance: int  # millisatoshis
-    total_deposited: int = 0
+class ApiKey(SQLModel, table=True):
+    hashed_key: str = Field(primary_key=True)
+    balance: int
+    reserved_balance: int = 0
+    refund_address: str | None = None
+    key_expiry_time: int | None = None
     total_spent: int = 0
-    created_at: datetime
-    expires_at: datetime | None = None
-    metadata: dict = Field(default_factory=dict, sa_column=Column(JSON))
+    total_requests: int = 0
 
-class Transaction(SQLModel, table=True):
-    id: int | None = Field(primary_key=True)
-    api_key_id: int = Field(foreign_key="apikey.id")
-    amount: int  # can be negative
-    balance_after: int
-    type: TransactionType
-    description: str
-    timestamp: datetime
+class LightningInvoice(SQLModel, table=True):
+    id: str = Field(primary_key=True)
+    bolt11: str
+    amount_sats: int
+    status: str
+
+class UpstreamProviderRow(SQLModel, table=True):
+    id: int | None = Field(default=None, primary_key=True)
+    provider_type: str
+    base_url: str
+    api_key: str
 ```
 
 ### Admin Interface
@@ -245,18 +219,17 @@ class Transaction(SQLModel, table=True):
 Web dashboard and admin API:
 
 ```python
-@admin_router.get("/admin/")
+@admin_router.get("/admin")
 async def admin_dashboard(request: Request):
     """Render admin HTML interface"""
     # Authentication check
     # Load statistics
     # Render template
 
-@admin_router.post("/admin/api/withdraw")
+@admin_router.post("/admin/withdraw")
 async def withdraw_balance(
-    api_key: str,
-    amount: int | None = None
-) -> WithdrawalResponse:
+    request: Request, withdraw_request: WithdrawRequest
+) -> dict[str, str]:
     """Generate eCash token for withdrawal"""
 ```
 
@@ -271,25 +244,14 @@ Features:
 
 #### `routstr/wallet.py`
 
-Cashu wallet operations:
+Cashu wallet operations (function-based):
 
 ```python
-class WalletManager:
-    """Manage Cashu wallet instances"""
-    
-    async def redeem_token(
-        self, 
-        token: str,
-        mint_url: str | None = None
-    ) -> int:
-        """Redeem eCash token and return value"""
-        
-    async def create_token(
-        self,
-        amount: int,
-        mint_url: str
-    ) -> str:
-        """Create eCash token for withdrawal"""
+async def recieve_token(token: str) -> tuple[int, str, str]:
+    """Redeem eCash token and return amount/unit/mint."""
+
+async def send_token(amount: int, unit: str, mint_url: str | None = None) -> str:
+    """Create eCash token for withdrawal."""
 ```
 
 ### Utility Modules
@@ -301,12 +263,9 @@ Structured logging configuration:
 ```python
 def setup_logging():
     """Configure JSON structured logging"""
-    # Set log level
-    # Configure formatters
-    # Add handlers
-    
-class RequestIdMiddleware:
-    """Add request ID to all logs"""
+
+class RequestIdFilter(logging.Filter):
+    """Attach request ID to log records."""
 ```
 
 #### `routstr/core/middleware.py`
@@ -316,27 +275,18 @@ HTTP middleware components:
 ```python
 class LoggingMiddleware:
     """Log all HTTP requests/responses"""
-    
-class ErrorHandlingMiddleware:
-    """Consistent error responses"""
 ```
 
 #### `routstr/core/exceptions.py`
 
-Custom exception hierarchy:
+Exception handlers:
 
 ```python
-class RoustrError(Exception):
-    """Base exception with error details"""
-    status_code: int
-    error_type: str
-    detail: str
+async def http_exception_handler(request: Request, exc: Exception) -> JSONResponse:
+    """HTTP exception handler with request ID"""
 
-class PaymentError(RoustrError):
-    """Payment-related errors"""
-
-class UpstreamError(RoustrError):
-    """Upstream API errors"""
+async def general_exception_handler(request: Request, exc: Exception) -> JSONResponse:
+    """Fallback exception handler with request ID"""
 ```
 
 ## Configuration Files
@@ -348,7 +298,7 @@ Project metadata and dependencies:
 ```toml
 [project]
 name = "routstr"
-version = "0.2.0"
+version = "0.2.2"
 dependencies = [
     "fastapi[standard]>=0.115",
     "sqlmodel>=0.0.24",
@@ -409,13 +359,13 @@ Using FastAPI's DI system:
 
 ```python
 # Define dependency
-async def get_db() -> AsyncSession:
-    async with async_session() as session:
+async def get_session() -> AsyncGenerator[AsyncSession, None]:
+    async with AsyncSession(engine, expire_on_commit=False) as session:
         yield session
 
 # Use in routes
 @router.get("/items")
-async def get_items(db: AsyncSession = Depends(get_db)):
+async def get_items(db: AsyncSession = Depends(get_session)):
     result = await db.execute(select(Item))
     return result.scalars().all()
 ```

docs/contributing/setup.md

@@ -22,22 +22,7 @@ git clone https://github.com/YOUR_USERNAME/routstr-core.git
 cd routstr-core
 ```
 
-### 2. Install uv
-
-We use [uv](https://github.com/astral-sh/uv) for fast, reliable Python package management:
-
-```bash
-# Using the installer script
-curl -LsSf https://astral.sh/uv/install.sh | sh
-
-# Or with pip
-pip install uv
-
-# Or with Homebrew (macOS)
-brew install uv
-```
-
-### 3. Set Up Environment
+### 2. Set Up Environment
 
 Run the setup command:
 
@@ -47,13 +32,13 @@ make setup
 
 This will:
 
-- ✅ Install uv if not present
+- ✅ Install [uv](https://github.com/astral-sh/uv) if not present
 - ✅ Create a virtual environment
 - ✅ Install all dependencies
 - ✅ Install dev tools (mypy, ruff, pytest)
 - ✅ Install project in editable mode
 
-### 4. Configure Environment
+### 3. Configure Environment
 
 Create your environment file:
 
@@ -71,7 +56,7 @@ ADMIN_PASSWORD=development-password
 DATABASE_URL=sqlite+aiosqlite:///dev.db
 ```
 
-### 5. Verify Installation
+### 4. Verify Installation
 
 Run these commands to verify your setup:
 
@@ -168,42 +153,92 @@ Understanding the codebase:
 ```
 routstr-core/
 ├── routstr/                    # Main package
-│   ├── __init__.py            # Package initialization
+│   ├── __init__.py
+│   ├── algorithm.py           # Provider selection algorithms
 │   ├── auth.py                # Authentication logic
-│   ├── balance.py             # Balance management
+│   ├── balance.py             # Balance management API
 │   ├── discovery.py           # Nostr discovery
+│   ├── lightning.py           # Lightning invoice handling
+│   ├── nip91.py               # Node announcement implementation
 │   ├── proxy.py               # Request proxying
 │   ├── wallet.py              # Cashu wallet integration
 │   │
 │   ├── core/                  # Core modules
-│   │   ├── admin.py          # Admin dashboard
-│   │   ├── db.py             # Database models
+│   │   ├── admin.py          # Admin dashboard API
+│   │   ├── db.py             # Database models (SQLModel)
 │   │   ├── exceptions.py     # Custom exceptions
+│   │   ├── log_manager.py    # Log management
 │   │   ├── logging.py        # Logging setup
-│   │   ├── main.py           # FastAPI app
-│   │   └── middleware.py     # HTTP middleware
+│   │   ├── main.py           # FastAPI app entry
+│   │   ├── middleware.py     # HTTP middleware
+│   │   └── settings.py       # Configuration
 │   │
-│   └── payment/               # Payment processing
-│       ├── cost_calculation.py # Cost logic
-│       ├── helpers.py         # Utilities
-│       ├── lnurl.py          # Lightning URLs
-│       ├── models.py         # Model pricing
-│       ├── price.py          # BTC pricing
-│       └── x_cashu.py        # Cashu headers
+│   ├── payment/               # Payment processing
+│   │   ├── cost_calculation.py
+│   │   ├── helpers.py
+│   │   ├── lnurl.py          # LNURL support
+│   │   ├── models.py         # Model pricing
+│   │   └── price.py          # BTC/USD rates
+│   │
+│   └── upstream/              # Upstream providers
+│       ├── base.py           # Base provider class
+│       ├── helpers.py        # Shared utilities
+│       ├── openai.py         # OpenAI
+│       ├── anthropic.py      # Anthropic
+│       ├── gemini.py         # Google Gemini
+│       ├── openrouter.py     # OpenRouter
+│       └── ...               # More providers
+│
+├── ui/                        # Admin dashboard (Next.js)
+│   ├── app/                  # Next.js app router
+│   │   ├── page.tsx         # Landing page
+│   │   ├── balances/        # Balance management
+│   │   ├── logs/            # Request logs viewer
+│   │   ├── model/           # Model configuration
+│   │   ├── providers/       # Upstream providers
+│   │   ├── settings/        # Node settings
+│   │   └── transactions/    # Transaction history
+│   ├── components/           # React components
+│   │   ├── ui/              # shadcn/ui primitives
+│   │   ├── landing/         # Landing page components
+│   │   └── settings/        # Settings components
+│   └── lib/                  # Utilities & API client
+│       ├── api/             # Backend API client
+│       ├── auth/            # Auth context
+│       └── hooks/           # React hooks
 │
 ├── tests/                     # Test suite
-│   ├── unit/                 # Unit tests
-│   └── integration/          # Integration tests
+├── migrations/                # Alembic migrations
+├── scripts/                   # Utility scripts
+├── docs/                      # Documentation
+├── examples/                  # Usage examples
 │
-├── migrations/               # Database migrations
-├── scripts/                  # Utility scripts
-├── docs/                     # Documentation
-│
-├── Makefile                  # Dev commands
-├── pyproject.toml           # Project config
-└── compose.yml              # Docker setup
+├── Makefile                   # Dev commands
+├── pyproject.toml             # Project config
+└── compose.yml                # Docker setup
 ```
 
+### Admin Dashboard (UI)
+
+The admin dashboard is a Next.js app using:
+
+- **Next.js 14** with App Router
+- **shadcn/ui** for components
+- **Tailwind CSS** for styling
+- **pnpm** for package management
+
+```bash
+# Development
+cd ui
+pnpm install
+pnpm dev          # http://localhost:3000
+
+# Build for production
+pnpm build
+```
+
+The UI is served by the FastAPI backend at `/admin/` when built. Use `make build-ui` to build and copy to the backend.
+
 ## Common Tasks
 
 ### Adding a New Endpoint
@@ -383,7 +418,7 @@ rm -rf test_*.db
 Now that you're set up:
 
 1. Read the [Architecture Overview](architecture.md)
-3. Check [open issues](https://github.com/routstr/routstr-core/issues)
-4. Start with a small contribution
+2. Check [open issues](https://github.com/routstr/routstr-core/issues)
+3. Start with a small contribution
 
 Happy coding! 🚀

docs/contributing/testing.md

@@ -6,576 +6,228 @@ This guide covers testing practices, patterns, and tools used in Routstr Core de
 
 We follow these principles:
 
-- **Test Behavior, Not Implementation** - Tests should survive refactoring
-- **Fast Feedback** - Unit tests run in milliseconds
-- **Reliable Tests** - No flaky tests allowed
-- **Clear Failures** - Tests should clearly indicate what broke
+- Test behavior, not implementation
+- Fast feedback
+- Reliable tests
+- Clear failures
 
 ## Test Structure
 
 ```
 tests/
-├── __init__.py
-├── conftest.py              # Shared fixtures and configuration
-├── unit/                    # Fast, isolated unit tests
-│   ├── test_auth.py
-│   ├── test_balance.py
-│   ├── test_cost_calculation.py
-│   ├── test_models.py
-│   └── test_wallet.py
-└── integration/             # Component integration tests
-    ├── test_api_endpoints.py
-    ├── test_payment_flow.py
-    ├── test_proxy_streaming.py
-    └── test_real_mint.py
+├── integration/
+│   ├── conftest.py
+│   ├── utils.py
+│   ├── test_wallet_topup.py
+│   ├── test_wallet_refund.py
+│   ├── test_wallet_information.py
+│   ├── test_proxy_get_endpoints.py
+│   ├── test_proxy_post_endpoints.py
+│   └── ... more integration tests
+├── unit/
+│   ├── test_algorithm.py
+│   ├── test_fee_consistency.py
+│   ├── test_image_tokens.py
+│   ├── test_logging_securityfilter.py
+│   ├── test_payment_helpers.py
+│   ├── test_settings.py
+│   ├── test_wallet.py
+│   └── ... more unit tests
+└── run_integration.py
 ```
 
 ## Running Tests
 
-### Quick Test Commands
+### Make Targets
+
+```bash
+# Run all tests (unit + integration with mocks)
+make test
+
+# Unit tests only
+make test-unit
+
+# Integration tests with mocks (fast)
+make test-integration
+
+# Integration tests with Docker services
+make test-integration-docker
+
+# Fast tests only (skip slow and Docker tests)
+make test-fast
+
+# Performance tests
+make test-performance
+
+# Coverage
+make test-coverage
+```
+
+### Direct pytest Commands
 
 ```bash
 # Run all tests
-make test
+pytest
 
-# Run unit tests only (fast)
-make test-unit
+# Run a specific test file
+pytest tests/unit/test_wallet.py -v
 
-# Run integration tests
-make test-integration
+# Run a specific test
+pytest tests/unit/test_wallet.py::test_get_balance -v
 
-# Run with coverage
-make test-coverage
-
-# Run specific test file
-uv run pytest tests/unit/test_auth.py -v
-
-# Run specific test
-uv run pytest tests/unit/test_auth.py::test_create_api_key -v
-
-# Run tests matching pattern
-uv run pytest -k "balance" -v
+# Run tests matching a pattern
+pytest -k "wallet" -v
 ```
 
-### Test Markers
+## Test Modes (Integration)
 
-Use markers to categorize tests:
+Integration tests support two execution modes:
 
-```python
-@pytest.mark.slow
-async def test_heavy_computation():
-    pass
+- Mock mode (default): uses in-memory mocks, no Docker required
+- Docker mode: uses real Docker services (Cashu mint, mock OpenAI, Nostr relay)
 
-@pytest.mark.requires_docker
-async def test_real_services():
-    pass
+Use the runner script for Docker mode:
 
-# Run without slow tests
-pytest -m "not slow"
+```bash
+./tests/run_integration.py
+```
+
+Or manually:
+
+```bash
+docker-compose -f compose.testing.yml up -d
+USE_LOCAL_SERVICES=1 pytest tests/integration/ -v
+docker-compose -f compose.testing.yml down -v
+```
+
+## Test Markers
+
+Markers are defined in `pyproject.toml`:
+
+- `integration`
+- `unit`
+- `slow`
+- `requires_docker`
+- `requires_real_mint`
+- `performance`
+- `asyncio`
+
+Examples:
+
+```bash
+# Skip slow tests
+pytest -m "not slow" -v
 
 # Run only integration tests
-pytest -m "integration"
+pytest -m "integration" -v
+
+# Run performance tests
+pytest -m "performance" -v
 ```
 
+## Fixtures and Utilities
+
+### Core Integration Fixtures
+
+Defined in `tests/integration/conftest.py`:
+
+- `integration_client` - Async HTTP client for the FastAPI app
+- `authenticated_client` - Client with a pre-created API key
+- `testmint_wallet` - Test wallet for generating Cashu tokens
+- `db_snapshot` - Database state snapshot/diff helper
+- `create_api_key` - Helper to create API keys for tests
+- `integration_engine`, `integration_session` - Async DB engine/session
+- `background_tasks_controller` - Control background tasks in tests
+- `mock_upstream_server` - Mock upstream API responses
+
+### Integration Utilities
+
+Defined in `tests/integration/utils.py`:
+
+- `CashuTokenGenerator`
+- `ResponseValidator`
+- `PerformanceValidator`
+- `ConcurrencyTester`
+- `DatabaseStateValidator`
+- `MockServiceBuilder`
+- `TestDataBuilder`
+
 ## Writing Tests
 
 ### Unit Test Example
 
 ```python
-# tests/unit/test_auth.py
-import pytest
-from routstr.auth import create_api_key, validate_api_key
+from routstr.algorithm import calculate_model_cost_score
+from routstr.payment.models import Architecture, Model, Pricing
 
-class TestAPIKeyAuth:
-    """Test API key authentication functionality"""
-    
-    async def test_create_api_key(self, test_db):
-        """Test creating a new API key"""
-        # Arrange
-        initial_balance = 10000
-        
-        # Act
-        api_key = await create_api_key(
-            balance=initial_balance,
-            name="Test Key"
-        )
-        
-        # Assert
-        assert api_key.key.startswith("sk-")
-        assert len(api_key.key) == 32
-        assert api_key.balance == initial_balance
-        
-    async def test_validate_invalid_key(self, test_db):
-        """Test validation fails for invalid key"""
-        # Act & Assert
-        with pytest.raises(InvalidAPIKeyError):
-            await validate_api_key("invalid_key")
+
+def test_calculate_model_cost_score_basic() -> None:
+    model = Model(
+        id="test-model",
+        name="Test test-model",
+        created=1234567890,
+        description="Test model",
+        context_length=8192,
+        architecture=Architecture(
+            modality="text",
+            input_modalities=["text"],
+            output_modalities=["text"],
+            tokenizer="gpt",
+            instruct_type=None,
+        ),
+        pricing=Pricing(
+            prompt=0.001,
+            completion=0.002,
+            request=0.0,
+            image=0.0,
+            web_search=0.0,
+            internal_reasoning=0.0,
+        ),
+    )
+    assert calculate_model_cost_score(model) == 0.002
 ```
 
 ### Integration Test Example
 
 ```python
-# tests/integration/test_payment_flow.py
 import pytest
 from httpx import AsyncClient
 
-class TestPaymentFlow:
-    """Test end-to-end payment flows"""
-    
-    @pytest.mark.asyncio
-    async def test_token_redemption_flow(
-        self, 
-        app_client: AsyncClient,
-        mock_cashu_wallet
-    ):
-        """Test complete token redemption and API usage"""
-        # Arrange
-        token = create_test_token(amount=5000)
-        mock_cashu_wallet.redeem.return_value = 5000
-        
-        # Act - Create API key
-        response = await app_client.post(
-            "/v1/wallet/create",
-            json={"cashu_token": token}
-        )
-        
-        # Assert - Key created
-        assert response.status_code == 200
-        api_key = response.json()["api_key"]
-        assert response.json()["balance"] == 5000
-        
-        # Act - Use API key
-        response = await app_client.post(
-            "/v1/chat/completions",
-            headers={"Authorization": f"Bearer {api_key}"},
-            json={
-                "model": "gpt-3.5-turbo",
-                "messages": [{"role": "user", "content": "Hi"}]
-            }
-        )
-        
-        # Assert - Request successful
-        assert response.status_code == 200
-        assert "choices" in response.json()
-```
 
-## Test Fixtures
-
-### Common Fixtures
-
-Located in `tests/conftest.py`:
-
-```python
-@pytest.fixture
-async def test_db():
-    """Provide a clean test database"""
-    # Create in-memory SQLite database
-    engine = create_async_engine("sqlite+aiosqlite:///:memory:")
-    
-    async with engine.begin() as conn:
-        await conn.run_sync(SQLModel.metadata.create_all)
-    
-    async_session = sessionmaker(engine, class_=AsyncSession)
-    
-    async with async_session() as session:
-        yield session
-    
-    await engine.dispose()
-
-@pytest.fixture
-async def app_client(test_db):
-    """Provide test client with test database"""
-    app.dependency_overrides[get_db] = lambda: test_db
-    
-    async with AsyncClient(app=app, base_url="http://test") as client:
-        yield client
-    
-    app.dependency_overrides.clear()
-
-@pytest.fixture
-def mock_cashu_wallet(mocker):
-    """Mock Cashu wallet for testing"""
-    mock = mocker.patch("routstr.wallet.Wallet")
-    mock.return_value.redeem.return_value = 1000
-    return mock
-```
-
-### Using Fixtures
-
-```python
-async def test_with_fixtures(
-    test_db,          # Get test database
-    app_client,       # Get test HTTP client
-    mock_cashu_wallet # Get mocked wallet
-):
-    # Use fixtures in test
-    pass
-```
-
-## Mocking Strategies
-
-### Mocking External Services
-
-```python
-# Mock upstream API
-@pytest.fixture
-def mock_openai(mocker):
-    mock_response = mocker.Mock()
-    mock_response.json.return_value = {
-        "choices": [{
-            "message": {"content": "Hello!"},
-            "finish_reason": "stop"
-        }],
-        "usage": {
-            "prompt_tokens": 10,
-            "completion_tokens": 20,
-            "total_tokens": 30
-        }
-    }
-    
-    mocker.patch(
-        "httpx.AsyncClient.post",
-        return_value=mock_response
-    )
-
-# Mock Cashu mint
-@pytest.fixture
-def mock_mint(mocker):
-    mint = mocker.patch("routstr.wallet.Mint")
-    mint.return_value.check_proof_state.return_value = True
-    return mint
-```
-
-### Mocking Time
-
-```python
-from freezegun import freeze_time
-
-@freeze_time("2024-01-01 12:00:00")
-async def test_time_dependent():
-    # Time is frozen during test
-    key = await create_api_key(expires_in_days=30)
-    assert key.expires_at == datetime(2024, 1, 31, 12, 0, 0)
-```
-
-## Test Patterns
-
-### Testing Async Code
-
-```python
-# Always mark async tests
+@pytest.mark.integration
 @pytest.mark.asyncio
-async def test_async_function():
-    result = await async_operation()
-    assert result == expected
-
-# Test async context managers
-async def test_async_context():
-    async with create_resource() as resource:
-        assert resource.is_active
+async def test_wallet_topup(
+    authenticated_client: AsyncClient,
+    testmint_wallet: object,
+    db_snapshot: object,
+) -> None:
+    await db_snapshot.capture()
+    token = await testmint_wallet.mint_tokens(1000)
+    response = await authenticated_client.post(
+        "/v1/wallet/topup", params={"cashu_token": token}
+    )
+    assert response.status_code == 200
+    diff = await db_snapshot.diff()
+    assert len(diff["api_keys"]["modified"]) == 1
 ```
 
-### Testing Exceptions
-
-```python
-# Test specific exception
-async def test_raises_specific_error():
-    with pytest.raises(InsufficientBalanceError) as exc_info:
-        await deduct_balance(api_key, amount=999999)
-    
-    assert "Insufficient balance" in str(exc_info.value)
-    assert exc_info.value.status_code == 402
-
-# Test exception details
-async def test_exception_details():
-    with pytest.raises(RoustrError) as exc_info:
-        await risky_operation()
-    
-    error = exc_info.value
-    assert error.error_type == "validation_error"
-    assert error.detail == "Invalid input"
-```
-
-### Testing Streaming Responses
-
-```python
-async def test_streaming_response():
-    """Test streaming chat completion"""
-    chunks = []
-    
-    async with app_client.stream(
-        "POST",
-        "/v1/chat/completions",
-        json={"model": "gpt-3.5-turbo", "stream": True}
-    ) as response:
-        async for line in response.aiter_lines():
-            if line.startswith("data: "):
-                chunks.append(json.loads(line[6:]))
-    
-    # Verify chunks
-    assert len(chunks) > 0
-    assert chunks[-1] == "[DONE]"
-```
-
-### Database Testing
-
-```python
-async def test_database_transaction(test_db):
-    """Test atomic transactions"""
-    async with test_db.begin():
-        # Create test data
-        api_key = APIKey(key_hash="test", balance=1000)
-        test_db.add(api_key)
-        await test_db.flush()
-        
-        # Test rollback
-        try:
-            async with test_db.begin_nested():
-                api_key.balance = -100  # Invalid
-                await test_db.flush()
-                raise ValueError("Rollback")
-        except ValueError:
-            pass
-        
-        # Verify rollback worked
-        await test_db.refresh(api_key)
-        assert api_key.balance == 1000
-```
-
-## Performance Testing
-
-### Benchmark Tests
-
-```python
-@pytest.mark.benchmark
-def test_performance(benchmark):
-    """Benchmark critical functions"""
-    result = benchmark(expensive_function, arg1, arg2)
-    assert result == expected
-
-# Run benchmarks
-pytest --benchmark-only
-```
-
-### Load Testing
-
-```python
-# tests/integration/test_load.py
-async def test_concurrent_requests(app_client):
-    """Test handling multiple concurrent requests"""
-    async def make_request(i):
-        response = await app_client.get(f"/test/{i}")
-        return response.status_code
-    
-    # Make 100 concurrent requests
-    tasks = [make_request(i) for i in range(100)]
-    results = await asyncio.gather(*tasks)
-    
-    # All should succeed
-    assert all(status == 200 for status in results)
-```
-
-## Test Data
-
-### Factories
-
-```python
-# tests/factories.py
-from datetime import datetime, timedelta
-
-def create_test_api_key(**kwargs):
-    """Factory for test API keys"""
-    defaults = {
-        "key": f"sk-test-{uuid4().hex[:8]}",
-        "balance": 10000,
-        "created_at": datetime.utcnow(),
-        "expires_at": datetime.utcnow() + timedelta(days=30)
-    }
-    defaults.update(kwargs)
-    return APIKey(**defaults)
-
-def create_test_token(amount: int = 1000, mint: str = None):
-    """Factory for test Cashu tokens"""
-    # Create valid test token structure
-    return base64.encode(...)
-```
-
-### Test Constants
-
-```python
-# tests/constants.py
-TEST_MODELS = {
-    "gpt-3.5-turbo": {
-        "prompt": 0.0015,
-        "completion": 0.002
-    },
-    "gpt-4": {
-        "prompt": 0.03,
-        "completion": 0.06
-    }
-}
-
-TEST_API_KEY = "sk-test-1234567890"
-TEST_MINT_URL = "https://testmint.example.com"
-```
-
-## Coverage
-
-### Running Coverage
+## Debugging Tips
 
 ```bash
-# Generate coverage report
-make test-coverage
+# Show print output
+pytest -s tests/unit/test_wallet.py
 
-# View HTML report
-open htmlcov/index.html
-
-# Coverage with specific tests
-pytest --cov=routstr --cov-report=html tests/unit/
-```
-
-### Coverage Configuration
-
-In `pyproject.toml`:
-
-```toml
-[tool.coverage.run]
-source = ["routstr"]
-omit = ["tests/*", "*/migrations/*"]
-
-[tool.coverage.report]
-exclude_lines = [
-    "pragma: no cover",
-    "def __repr__",
-    "raise AssertionError",
-    "raise NotImplementedError",
-    "if TYPE_CHECKING:"
-]
-```
-
-## Debugging Tests
-
-### Print Debugging
-
-```python
-# Use -s flag to see print output
-pytest -s tests/unit/test_auth.py
-
-# In test
-async def test_debug():
-    print(f"Value: {value}")  # Will show with -s
-    assert value == expected
-```
-
-### Interactive Debugging
-
-```python
 # Drop into debugger on failure
 pytest --pdb
-
-# Set breakpoint in test
-async def test_debug():
-    import pdb; pdb.set_trace()
-    # Execution stops here
 ```
 
-### Logging in Tests
-
-```python
-# Enable debug logging in tests
-import logging
-logging.basicConfig(level=logging.DEBUG)
-
-# Or use caplog fixture
-async def test_logging(caplog):
-    with caplog.at_level(logging.INFO):
-        await function_that_logs()
-    
-    assert "Expected message" in caplog.text
-```
-
-## CI/CD Integration
-
-### GitHub Actions
-
-Tests run automatically on:
-
-- Pull requests
-- Pushes to main
-- Nightly schedules
-
-See `.github/workflows/test.yml` for configuration.
-
-### Pre-commit Hooks
-
-Install pre-commit hooks:
-
-```bash
-pre-commit install
-
-# Run manually
-pre-commit run --all-files
-```
-
-## Best Practices
-
-### Do's
-
-1. ✅ Write tests first (TDD)
-2. ✅ Keep tests simple and focused
-3. ✅ Use descriptive test names
-4. ✅ Test edge cases
-5. ✅ Mock external dependencies
-6. ✅ Use fixtures for setup
-7. ✅ Assert specific values
-
-### Don'ts
-
-1. ❌ Don't test implementation details
-2. ❌ Don't use production services
-3. ❌ Don't rely on test order
-4. ❌ Don't ignore flaky tests
-5. ❌ Don't skip error cases
-6. ❌ Don't use hard-coded waits
-7. ❌ Don't commit commented tests
-
 ## Troubleshooting
 
-### Common Issues
-
-**Async Test Errors**
-
-```python
-# Wrong
-def test_async():  # Missing async
-    await function()
-
-# Right
-async def test_async():
-    await function()
-```
-
-**Database State**
-
-```python
-# Ensure clean state
-@pytest.fixture(autouse=True)
-async def cleanup(test_db):
-    yield
-    # Cleanup after each test
-    await test_db.execute("DELETE FROM apikey")
-    await test_db.commit()
-```
-
-**Mock Not Working**
-
-```python
-# Check import path
-mocker.patch("routstr.wallet.Wallet")  # Full path
-# Not just "Wallet"
-```
+- Docker mode failures: check `docker ps` and `docker-compose -f compose.testing.yml logs`
+- Connection errors: make sure ports 3338, 3000, 8000, and 8088 are free
+- Slow tests: use `pytest -m "not slow"` or `make test-fast`
 
 ## Next Steps
 
-- See [Architecture](architecture.md) for system design
-- Read [Setup Guide](setup.md) for environment setup
+- See [Architecture](architecture.md)
+- Read [Setup Guide](setup.md)

docs/getting-started/configuration.md

@@ -1,260 +0,0 @@
-# Configuration
-
-Routstr Core is configured via a single settings row in the database. Environment variables are only used on first run to seed that row (with a few computed defaults like `ONION_URL`). After that, the database is the source of truth. You can update settings at runtime via the admin API. `DATABASE_URL` is always env-only.
-
-## Environment Variables
-
-### Core Settings
-
-| Variable | Description | Default | Required |
-|----------|-------------|---------|----------|
-| `UPSTREAM_BASE_URL` | Base URL of the OpenAI-compatible API to proxy | - | ✅ |
-| `UPSTREAM_API_KEY` | API key for the upstream service | - | ❌ |
-| `ADMIN_PASSWORD` | Password for admin dashboard access | - | ⚠️ |
-
-### Node Information
-
-| Variable | Description | Default | Required |
-|----------|-------------|---------|----------|
-| `NAME` | Public name of your Routstr node | `ARoutstrNode` | ❌ |
-| `DESCRIPTION` | Description of your node | `A Routstr Node` | ❌ |
-| `NPUB` | Nostr public key for node identity | - | ❌ |
-| `HTTP_URL` | Public HTTP URL of your node | - | ❌ |
-| `ONION_URL` | Tor hidden service URL | - | ❌ |
-
-### Cashu Configuration
-
-| Variable | Description | Default | Required |
-|----------|-------------|---------|----------|
-| `CASHU_MINTS` | Comma-separated list of trusted Cashu mint URLs | `https://mint.minibits.cash/Bitcoin` | ❌ |
-| `RECEIVE_LN_ADDRESS` | Lightning address for automatic payouts | - | ❌ |
-
-### Pricing Configuration
-
-| Variable | Description | Default | Required |
-|----------|-------------|---------|----------|
-| `FIXED_PRICING` | Force fixed per-request pricing (ignore model token pricing) | `false` | ❌ |
-| `FIXED_COST_PER_REQUEST` | Fixed cost per API request in sats | `1` | ❌ |
-| `FIXED_PER_1K_INPUT_TOKENS` | Optional override: sats per 1000 input tokens | `0` | ❌ |
-| `FIXED_PER_1K_OUTPUT_TOKENS` | Optional override: sats per 1000 output tokens | `0` | ❌ |
-| `EXCHANGE_FEE` | Exchange rate markup (1.005 = 0.5% fee) | `1.005` | ❌ |
-| `UPSTREAM_PROVIDER_FEE` | Provider fee markup (1.05 = 5% fee) | `1.05` | ❌ |
-
-### Network & Discovery
-
-| Variable | Description | Default | Required |
-|----------|-------------|---------|----------|
-| `CORS_ORIGINS` | Comma-separated list of allowed CORS origins | `*` | ❌ |
-| `TOR_PROXY_URL` | SOCKS5 proxy URL for Tor connections | `socks5://127.0.0.1:9050` | ❌ |
-| `RELAYS` | Comma-separated nostr relays used for provider discovery | sane defaults | ❌ |
-| `PROVIDERS_REFRESH_INTERVAL_SECONDS` | Provider cache refresh interval | `300` | ❌ |
-
-### Logging Configuration
-
-| Variable | Description | Default | Required |
-|----------|-------------|---------|----------|
-| `LOG_LEVEL` | Logging level (DEBUG, INFO, WARNING, ERROR) | `INFO` | ❌ |
-| `ENABLE_CONSOLE_LOGGING` | Enable console log output | `true` | ❌ |
-
-### Other
-
-| Variable | Description | Default | Required |
-|----------|-------------|---------|----------|
-| `CHAT_COMPLETIONS_API_VERSION` | Append `api-version` to `/chat/completions` (Azure OpenAI) | - | ❌ |
-| `DATABASE_URL` | SQLite database connection string | `sqlite+aiosqlite:///keys.db` | ❌ |
-| `REFUND_CACHE_TTL_SECONDS` | Cache TTL for refund responses (seconds) | `3600` | ❌ |
-
-## Configuration Examples
-
-### Basic OpenAI Proxy
-
-```bash
-# .env
-UPSTREAM_BASE_URL=https://api.openai.com/v1
-UPSTREAM_API_KEY=sk-...
-ADMIN_PASSWORD=my-secure-password
-```
-
-### Custom AI Provider
-
-```bash
-# .env
-UPSTREAM_BASE_URL=https://api.anthropic.com/v1
-UPSTREAM_API_KEY=your-anthropic-key
-MODELS_PATH=/app/config/anthropic-models.json
-```
-
-### Azure OpenAI (optional)
-
-```bash
-# .env
-UPSTREAM_BASE_URL=https://<resource>.openai.azure.com/openai/deployments/<deployment>
-UPSTREAM_API_KEY=<azure_api_key>
-CHAT_COMPLETIONS_API_VERSION=2024-05-01-preview
-```
-
-### High-Security Setup
-
-```bash
-# .env
-UPSTREAM_BASE_URL=https://api.openai.com/v1
-UPSTREAM_API_KEY=sk-...
-ADMIN_PASSWORD=very-long-secure-password-here
-CORS_ORIGINS=https://myapp.com,https://app.myapp.com
-TOR_PROXY_URL=socks5://tor:9050
-LOG_LEVEL=WARNING
-```
-
-### Public Node Configuration
-
-```bash
-# .env
-NAME=Lightning AI Gateway
-DESCRIPTION=Fast and reliable AI API access with Bitcoin payments
-NPUB=npub1abcd...
-HTTP_URL=https://api.lightning-ai.com
-ONION_URL=http://lightningai.onion
-CASHU_MINTS=https://mint1.com,https://mint2.com
-```
-
-## Pricing
-
-- Default: pricing comes from your `models.json`.
-- Force fixed per-request pricing: set `FIXED_PRICING=true` and `FIXED_COST_PER_REQUEST`.
-- Optional token overrides when using model pricing: set
-  `FIXED_PER_1K_INPUT_TOKENS` and/or `FIXED_PER_1K_OUTPUT_TOKENS`.
-- Legacy envs are still accepted and mapped automatically:
-  `MODEL_BASED_PRICING` → `!FIXED_PRICING`, `COST_PER_REQUEST` → `FIXED_COST_PER_REQUEST`,
-  `COST_PER_1K_*` → `FIXED_PER_1K_*`.
-
-Example fixed pricing:
-
-```bash
-FIXED_PRICING=true
-FIXED_COST_PER_REQUEST=10
-```
-
-## Custom Models Configuration
-
-Create a `models.json` file:
-
-```json
-{
-  "models": [
-    {
-      "id": "gpt-4",
-      "name": "GPT-4",
-      "pricing": {
-        "prompt": "0.00003",
-        "completion": "0.00006",
-        "request": "0"
-      }
-    },
-    {
-      "id": "gpt-3.5-turbo",
-      "name": "GPT-3.5 Turbo",
-      "pricing": {
-        "prompt": "0.0000015",
-        "completion": "0.000002",
-        "request": "0"
-      }
-    }
-  ]
-}
-```
-
-## Security Best Practices
-
-### Admin Password
-
-Generate a strong password:
-
-```bash
-openssl rand -base64 32
-```
-
-### API Keys
-
-- Rotate upstream API keys regularly
-- Use read-only keys when possible
-- Monitor key usage
-
-### Network Security
-
-- Restrict CORS origins in production
-- Use HTTPS for public endpoints
-- Enable Tor for anonymity
-
-### Database Security
-
-- Regular backups
-- Encrypted storage volumes
-- Restricted file permissions
-
-## Troubleshooting
-
-### Check Current Configuration
-
-```bash
-# View all environment variables
-docker exec routstr env | sort
-
-# Test configuration
-curl http://localhost:8000/v1/info
-```
-
-### Common Issues
-
-**Missing Upstream URL**
-
-```
-ERROR: UPSTREAM_BASE_URL not set
-Solution: Set UPSTREAM_BASE_URL in .env
-```
-
-**Invalid Cashu Mint**
-
-```
-ERROR: Failed to connect to mint
-Solution: Verify CASHU_MINTS URLs are accessible
-```
-
-**Database Errors**
-
-```
-ERROR: Database connection failed
-Solution: Check DATABASE_URL and file permissions
-```
-
-## Advanced Configuration
-
-### Multiple Mints
-
-Configure fallback mints:
-
-```bash
-CASHU_MINTS=https://primary.mint,https://backup1.mint,https://backup2.mint
-```
-
-### Custom Database
-
-Use PostgreSQL instead of SQLite:
-
-```bash
-DATABASE_URL=postgresql+asyncpg://user:pass@localhost/routstr
-```
-
-### Proxy Settings
-
-For corporate environments:
-
-```bash
-HTTP_PROXY=http://proxy.company.com:8080
-HTTPS_PROXY=http://proxy.company.com:8080
-```
-
-## Next Steps
-
-- [User Guide](../user-guide/introduction.md) - Start using Routstr
-- [Admin Dashboard](../user-guide/admin-dashboard.md) - Manage your node
-- [Custom Pricing](../advanced/custom-pricing.md) - Advanced pricing strategies

docs/getting-started/docker.md

@@ -1,337 +0,0 @@
-# Docker Setup
-
-This guide covers deploying Routstr Core using Docker for production environments.
-
-## Docker Images
-
-Official images are available on GitHub Container Registry:
-
-```bash
-ghcr.io/routstr/proxy:latest
-```
-
-## Basic Docker Run
-
-### Minimal Setup
-
-```bash
-docker run -d \
-  --name routstr \
-  -p 8000:8000 \
-  -e UPSTREAM_BASE_URL=https://api.openai.com/v1 \
-  -e UPSTREAM_API_KEY=sk-... \
-  -e ADMIN_PASSWORD=secure-password \
-  ghcr.io/routstr/proxy:latest
-```
-
-### With Persistent Storage
-
-```bash
-docker run -d \
-  --name routstr \
-  -p 8000:8000 \
-  -v routstr-data:/app/data \
-  -v routstr-logs:/app/logs \
-  -e UPSTREAM_BASE_URL=https://api.openai.com/v1 \
-  -e UPSTREAM_API_KEY=sk-... \
-  -e DATABASE_URL=sqlite+aiosqlite:///data/keys.db \
-  ghcr.io/routstr/proxy:latest
-```
-
-## Docker Compose Setup
-
-### Basic Configuration
-
-Create `compose.yml`:
-
-```yaml
-version: '3.8'
-
-services:
-  routstr:
-    image: ghcr.io/routstr/proxy:latest
-    ports:
-      - "8000:8000"
-    volumes:
-      - ./data:/app/data
-      - ./logs:/app/logs
-    env_file:
-      - .env
-    restart: unless-stopped
-```
-
-### With Tor Hidden Service
-
-The included `compose.yml` provides Tor support:
-
-```yaml
-version: '3.8'
-
-services:
-  routstr:
-    build: .  # Or use image: ghcr.io/routstr/proxy:latest
-    volumes:
-      - .:/app
-      - ./logs:/app/logs
-    env_file:
-      - .env
-    environment:
-      - TOR_PROXY_URL=socks5://tor:9050
-    ports:
-      - 8000:8000
-    extra_hosts:
-      - "host.docker.internal:host-gateway"
-
-  tor:
-    image: ghcr.io/hundehausen/tor-hidden-service:latest
-    volumes:
-      - tor-data:/var/lib/tor
-    environment:
-      - HS_ROUTER=routstr:8000:80
-    depends_on:
-      - routstr
-
-volumes:
-  tor-data:
-```
-
-### Environment File
-
-Create `.env` file:
-
-```bash
-# Required
-UPSTREAM_BASE_URL=https://api.openai.com/v1
-UPSTREAM_API_KEY=your-api-key
-ADMIN_PASSWORD=secure-admin-password
-
-# Cashu Configuration
-CASHU_MINTS=https://mint.minibits.cash/Bitcoin
-
-# Optional
-NAME=My Routstr Node
-DESCRIPTION=Pay-per-use AI API proxy
-NPUB=npub1...
-HTTP_URL=https://api.mynode.com
-ONION_URL=http://mynode.onion
-
-# Pricing (optional)
-FIXED_PRICING=false
-EXCHANGE_FEE=1.005
-UPSTREAM_PROVIDER_FEE=1.05
-```
-
-## Building Custom Image
-
-### Dockerfile Overview
-
-The provided Dockerfile:
-
-- Uses Alpine Linux for small size
-- Installs required dependencies for secp256k1
-- Runs as non-root user
-- Exposes port 8000
-
-### Build Locally
-
-```bash
-# Clone repository
-git clone https://github.com/routstr/routstr-core.git
-cd routstr-core
-
-# Build image
-docker build -t my-routstr:latest .
-
-# Run custom image
-docker run -d \
-  --name routstr \
-  -p 8000:8000 \
-  --env-file .env \
-  my-routstr:latest
-```
-
-## Deployment Considerations
-
-### Resource Requirements
-
-- **CPU**: 1-2 cores recommended
-- **Memory**: 512MB-1GB
-- **Storage**: 1GB + database growth
-- **Network**: Low latency to upstream provider
-
-### Health Checks
-
-Add health check to compose.yml:
-
-```yaml
-services:
-  routstr:
-    # ... other config ...
-    healthcheck:
-      test: ["CMD", "curl", "-f", "http://localhost:8000/v1/info"]
-      interval: 30s
-      timeout: 10s
-      retries: 3
-      start_period: 40s
-```
-
-### Reverse Proxy Setup
-
-#### Nginx Example
-
-```nginx
-server {
-    listen 443 ssl http2;
-    server_name api.yournode.com;
-
-    ssl_certificate /path/to/cert.pem;
-    ssl_certificate_key /path/to/key.pem;
-
-    location / {
-        proxy_pass http://localhost:8000;
-        proxy_set_header Host $host;
-        proxy_set_header X-Real-IP $remote_addr;
-        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
-        proxy_set_header X-Forwarded-Proto $scheme;
-        
-        # For streaming responses
-        proxy_buffering off;
-        proxy_cache off;
-        proxy_set_header Connection '';
-        proxy_http_version 1.1;
-        chunked_transfer_encoding off;
-    }
-}
-```
-
-#### Caddy Example
-
-```caddy
-api.yournode.com {
-    reverse_proxy localhost:8000 {
-        flush_interval -1
-    }
-}
-```
-
-## Monitoring
-
-### Log Management
-
-View logs:
-
-```bash
-# Docker
-docker logs -f routstr
-
-# Docker Compose
-docker compose logs -f routstr
-
-# Log files
-tail -f ./logs/routstr.log
-```
-
-### Metrics
-
-Monitor key metrics:
-
-- Request count and latency
-- Token validation success rate
-- Upstream API errors
-- Database size growth
-
-## Backup and Recovery
-
-### Database Backup
-
-```bash
-# Backup SQLite database
-docker exec routstr sqlite3 /app/data/keys.db ".backup /app/data/backup.db"
-
-# Copy backup locally
-docker cp routstr:/app/data/backup.db ./backup-$(date +%Y%m%d).db
-```
-
-### Restore from Backup
-
-```bash
-# Stop service
-docker compose down
-
-# Restore database
-docker cp ./backup.db routstr:/app/data/keys.db
-
-# Restart service
-docker compose up -d
-```
-
-## Security Considerations
-
-### Environment Variables
-
-- Never commit `.env` files
-- Use Docker secrets for sensitive data
-- Rotate API keys regularly
-- Use strong admin passwords
-
-### Network Security
-
-- Use HTTPS/TLS termination
-- Restrict admin interface access
-- Enable firewall rules
-- Monitor for suspicious activity
-
-### Container Security
-
-- Run as non-root user
-- Use read-only filesystem where possible
-- Limit container capabilities
-- Keep base image updated
-
-## Troubleshooting
-
-### Container Won't Start
-
-```bash
-# Check logs
-docker logs routstr
-
-# Verify environment
-docker exec routstr env | grep -E "(UPSTREAM|CASHU|ADMIN)"
-
-# Test database connection
-docker exec routstr sqlite3 /app/data/keys.db ".tables"
-```
-
-### Permission Issues
-
-```bash
-# Fix volume permissions
-sudo chown -R 1000:1000 ./data ./logs
-```
-
-### Network Issues
-
-```bash
-# Test upstream connectivity
-docker exec routstr curl -I https://api.openai.com
-
-# Check DNS resolution
-docker exec routstr nslookup api.openai.com
-```
-
-## Production Checklist
-
-- [ ] Set strong `ADMIN_PASSWORD`
-- [ ] Configure proper `UPSTREAM_BASE_URL` and `UPSTREAM_API_KEY`
-- [ ] Set up persistent volumes for data and logs
-- [ ] Configure reverse proxy with TLS
-- [ ] Set up monitoring and alerting
-- [ ] Implement backup strategy
-- [ ] Test disaster recovery
-- [ ] Document deployment process
-
-## Next Steps
-
-- [Configuration Guide](configuration.md) - All environment variables
-- [Admin Dashboard](../user-guide/admin-dashboard.md) - Manage your node

docs/getting-started/overview.md

@@ -1,156 +0,0 @@
-# Overview
-
-Routstr Core is a powerful payment proxy that brings Bitcoin micropayments to AI APIs. This overview will help you understand the core concepts and architecture.
-
-## Core Concepts
-
-### Payment Proxy
-
-Routstr acts as a transparent proxy between your application and OpenAI-compatible APIs. It:
-
-- Intercepts API requests
-- Validates payment tokens
-- Forwards requests to the upstream provider
-- Tracks usage and deducts costs
-- Returns responses to the client
-
-### Cashu eCash Protocol
-
-[Cashu](https://cashu.space) is a Bitcoin eCash protocol that enables:
-
-- **Privacy**: Payments are unlinkable and untraceable
-- **Instant Settlement**: No waiting for blockchain confirmations
-- **Micropayments**: Send fractions of a satoshi
-- **Offline Capability**: Tokens can be transferred without internet
-
-### Lightning Network Integration
-
-Routstr connects to the Lightning Network through Cashu mints, enabling:
-
-- Fast Bitcoin deposits and withdrawals
-- Global payment reach
-- Low transaction fees
-- No minimum payment amounts
-
-## Architecture
-
-### System Components
-
-```mermaid
-graph TB
-    subgraph "Client Side"
-        A[AI Application]
-        B[OpenAI SDK]
-        C[eCash Wallet]
-    end
-    
-    subgraph "Routstr Core"
-        D[FastAPI Server]
-        E[Auth Module]
-        F[Payment Module]
-        G[Proxy Module]
-        H[SQLite Database]
-    end
-    
-    subgraph "External Services"
-        I[Upstream AI Provider]
-        J[Cashu Mint]
-        K[Bitcoin/Lightning]
-    end
-    
-    A --> B
-    B --> D
-    C --> D
-    D --> E
-    E --> F
-    F --> J
-    D --> G
-    G --> I
-    D --> H
-    J --> K
-```
-
-### Key Modules
-
-1. **Authentication** (`auth.py`)
-   - API key validation
-   - Balance checking
-   - Request authorization
-
-2. **Payment Processing** (`payment/`)
-   - Token validation
-   - Cost calculation
-   - Balance updates
-   - Pricing models
-
-3. **Proxy Handler** (`proxy.py`)
-   - Request forwarding
-   - Response streaming
-   - Usage tracking
-   - Error handling
-
-4. **Wallet Management** (`wallet.py`)
-   - Cashu wallet integration
-   - Token redemption
-   - Balance management
-   - Automatic payouts
-
-5. **Admin Interface** (`core/admin.py`)
-   - Web dashboard
-   - Balance viewing
-   - Key management
-   - Withdrawal interface
-
-## Payment Flow
-
-### Standard Flow (API Key)
-
-1. User deposits eCash tokens to create an API key
-2. Client sends requests with the API key
-3. Routstr checks balance and forwards request
-4. Cost is deducted based on actual usage
-5. Response is returned to client
-
-### Per-Request Flow (Coming Soon)
-
-1. Client includes eCash token in request header
-2. Routstr validates token meets minimum amount
-3. Request is processed
-4. Change is returned in response header
-5. No account or balance needed
-
-## Supported Features
-
-### API Compatibility
-
-- ✅ Chat completions (streaming and non-streaming)
-- ✅ Text completions
-- ✅ Embeddings
-- ✅ Image generation
-- ✅ Audio transcription/translation
-- ✅ Model listing
-- ✅ Custom endpoints
-
-### Payment Features
-
-- ✅ Multiple Cashu mint support
-- ✅ Automatic balance tracking
-- ✅ Model-based pricing
-- ✅ USD to BTC conversion
-- ✅ Configurable fees
-- ✅ Balance withdrawals
-
-### Operational Features
-
-- ✅ Docker deployment
-- ✅ Tor hidden service support
-- ✅ Nostr relay discovery
-- ✅ Database migrations
-- ✅ Comprehensive logging
-- ✅ Admin dashboard
-
-## Next Steps
-
-- [Quick Start](quickstart.md) - Get running in minutes
-- [Docker Setup](docker.md) - Deploy with containers
-- [Configuration](configuration.md) - Customize your instance

docs/getting-started/quickstart.md

@@ -1,226 +0,0 @@
-# Quick Start
-
-Get Routstr Core up and running in minutes with Docker or local development setup.
-
-## Prerequisites
-
-- Docker and Docker Compose (for production)
-- Python 3.11+ (for development)
-- A Cashu-compatible wallet (optional for testing)
-
-## Option 1: Docker (Recommended)
-
-### Quick Run
-
-The fastest way to start Routstr Core:
-
-```bash
-docker run -d \
-  --name routstr-proxy \
-  -p 8000:8000 \
-  -e UPSTREAM_BASE_URL=https://api.openai.com/v1 \
-  -e UPSTREAM_API_KEY=your-openai-api-key \
-  ghcr.io/routstr/proxy:latest
-```
-
-### Docker Compose
-
-For a full setup with Tor support:
-
-1. Clone the repository:
-
-```bash
-git clone https://github.com/routstr/routstr-core.git
-cd routstr-core
-```
-
-2. Create environment file:
-
-```bash
-cp .env.example .env
-# Edit .env with your settings
-```
-
-3. Start the services:
-
-```bash
-docker compose up -d
-```
-
-This will start:
-
-- Routstr proxy on port 8000
-- Tor hidden service (optional)
-- Automatic database migrations
-
-### Verify Installation
-
-Check that Routstr is running:
-
-```bash
-curl http://localhost:8000/v1/info
-```
-
-You should see:
-
-```json
-{
-  "name": "ARoutstrNode",
-  "description": "A Routstr Node",
-  "version": "0.2.0",
-  "npub": "",
-  "mints": ["https://mint.minibits.cash/Bitcoin"],
-  "models": {...}
-}
-```
-
-## Option 2: Local Development
-
-### Install Dependencies
-
-1. Install [uv](https://github.com/astral-sh/uv) package manager:
-
-```bash
-curl -LsSf https://astral.sh/uv/install.sh | sh
-```
-
-2. Clone and setup:
-
-```bash
-git clone https://github.com/routstr/routstr-core.git
-cd routstr-core
-uv sync
-```
-
-3. Configure environment:
-
-```bash
-cp .env.example .env
-# Edit .env with your settings
-```
-
-### Run the Server
-
-```bash
-fastapi run routstr --host 0.0.0.0 --port 8000
-```
-
-## First API Call
-
-### 1. Get an eCash Token
-
-You'll need a Cashu token to pay for API calls. Options:
-
-- Use a [Cashu wallet](https://cashu.space) to create tokens
-- Get test tokens from a testnet mint
-- Use the example token (for testing only)
-
-### 2. Create an API Key
-
-Send your eCash token to create an API key:
-
-```bash
-curl -X POST http://localhost:8000/v1/wallet/create \
-  -H "Content-Type: application/json" \
-  -d '{
-    "cashu_token": "cashuAeyJ0b2..."
-  }'
-```
-
-Response:
-
-```json
-{
-  "api_key": "rUvK7...",
-  "balance": 10000
-}
-```
-
-### 3. Make an API Call
-
-Use your API key like a normal OpenAI key:
-
-```python
-import openai
-
-client = openai.OpenAI(
-    api_key="rUvK7...",  # Your Routstr API key
-    base_url="http://localhost:8000/v1"
-)
-
-response = client.chat.completions.create(
-    model="gpt-4o-mini",
-    messages=[{"role": "user", "content": "Hello!"}]
-)
-
-print(response.choices[0].message.content)
-```
-
-## Example Client
-
-Run the included example:
-
-```bash
-CASHU_TOKEN="your-token" python example.py
-```
-
-This demonstrates:
-
-- Creating an API key from a token
-- Making streaming chat requests
-- Automatic balance deduction
-
-## Testing the Setup
-
-### Check Available Models
-
-```bash
-curl http://localhost:8000/v1/models
-```
-
-### View Admin Dashboard
-
-Open <http://localhost:8000/admin/> in your browser.
-
-Default password is set in `ADMIN_PASSWORD` environment variable.
-
-### Monitor Logs
-
-Docker:
-
-```bash
-docker compose logs -f routstr
-```
-
-Local:
-
-```bash
-# Logs are in ./logs/ directory
-tail -f logs/routstr.log
-```
-
-## Common Issues
-
-### Connection Refused
-
-- Ensure the service is running: `docker ps`
-- Check firewall settings
-- Verify port 8000 is not in use
-
-### Invalid API Key
-
-- Ensure you've created an API key with sufficient balance
-- Check the token was valid and had value
-- Verify the mint URL is accessible
-
-### Upstream Errors
-
-- Check `UPSTREAM_BASE_URL` is correct
-- Verify `UPSTREAM_API_KEY` if required
-- Test upstream service directly
-
-## Next Steps
-
-- [Configuration Guide](configuration.md) - Customize settings
-- [Docker Setup](docker.md) - Production deployment
-- [User Guide](../user-guide/introduction.md) - Detailed usage

docs/getting-started/ui-configuration.md

@@ -1,53 +0,0 @@
-# UI Configuration
-
-This guide explains how to configure the Routstr UI for different environments.
-
-## Environment Variables
-
-The UI uses Next.js environment variables to configure API endpoints and authentication.
-
-### Centralized Configuration
-
-This project uses a centralized configuration approach with a single `.env` file in the project root. This file contains both backend and frontend configuration variables.
-
-Create or update your `.env` file in the project root:
-
-```bash
-# .env (in project root)
-
-# UI Configuration (NEXT_PUBLIC_ variables are exposed to the browser)
-NEXT_PUBLIC_API_URL=http://127.0.0.1:8000
-```
-
-### Development vs Production
-
-The same `.env` file is used for both development and production. Simply change the values:
-
-**Development:**
-
-```bash
-NEXT_PUBLIC_API_URL=http://127.0.0.1:8000
-```
-
-**Production:**
-
-```bash
-NEXT_PUBLIC_API_URL=https://api.yourroutstr.com
-```
-
-## Building the UI
-
-The build process automatically reads configuration from the root `.env` file:
-
-```bash
-# From the project root
-make ui-build
-# or
-./scripts/build-ui.sh
-```
-
-The build script will automatically:
-
-- Load `NEXT_PUBLIC_*` variables from the root `.env` file
-- Use them during the Next.js build process
-- Display warnings if the `.env` file is missing

docs/index.md

@@ -1,83 +1,44 @@
 # Routstr Core Documentation
 
-Welcome to the official documentation for **Routstr Core** - a FastAPI-based reverse proxy that enables Bitcoin micropayments for OpenAI-compatible APIs using the Cashu eCash protocol.
+**Routstr** is a decentralized protocol for permissionless AI inference. It enables an open marketplace where anyone can buy and sell compute using **Bitcoin eCash (Cashu)**.
 
-## What is Routstr Core?
+---
 
-Routstr Core is a payment proxy that sits between API clients and OpenAI-compatible services. It enables:
+## 🐣 For Clients (Users & Builders)
 
-- **Pay-per-request billing** using Bitcoin eCash tokens
-- **Seamless integration** with existing OpenAI clients
-- **Privacy-preserving payments** through the Cashu protocol
-- **Flexible pricing models** with per-token or per-request billing
-- **Multi-provider support** for various AI model providers
+If you want to use AI models in your application without accounts or KYC.
 
-### Key Features
+- **[Introduction](client/introduction.md)**: How the ecosystem works.
+- **[Payment Flow](client/payments.md)**: Funding sessions, topping up, and refunds.
+- **[Integration Guide](client/integration.md)**: Code examples for Python, JS, and cURL.
 
-- 🪙 **Cashu Wallet Integration** - Accept Lightning payments and redeem eCash tokens
-- 🔑 **API Key Management** - Secure key storage with balance tracking
-- 💰 **Dynamic Pricing** - Model-based pricing with live BTC/USD conversion
-- 🎛️ **Admin Dashboard** - Web interface for balance and key management
-- 🌐 **Nostr Discovery** - Find providers through decentralized relay network
-- 🐋 **Docker Support** - Easy deployment with optional Tor hidden service
-- ⚡ **Lightning Fast** - Minimal latency overhead for API requests
+## 🦁 For Providers (Node Operators)
 
-## How It Works
+If you want to run a node, resell API access, or monetize hardware.
 
-```mermaid
-sequenceDiagram
-    participant Client
-    participant Routstr as Routstr Proxy
-    participant DB as Database
-    participant Upstream as AI Provider
-    participant Wallet as Cashu Wallet
+- **[Quick Start](provider/quickstart.md)**: Deploy a node in 5 minutes.
+- **[Deployment](provider/deployment.md)**: Production Docker setup.
+- **[Configuration](provider/configuration.md)**: Environment variables and settings.
+- **[Dashboard](provider/dashboard.md)**: Managing your node visually.
+- **[Pricing Strategy](provider/pricing.md)**: Setting margins and fees.
+- **[Discovery](provider/discovery.md)**: Announcing your node on Nostr.
+- **[Tor Support](provider/tor.md)**: Running an anonymous hidden service.
 
-    Client->>Routstr: API Request + eCash Token
-    Routstr->>Wallet: Validate & Redeem Token
-    Wallet-->>Routstr: Token Value (sats)
-    Routstr->>DB: Store/Update Balance
-    Routstr->>Upstream: Forward API Request
-    Upstream-->>Routstr: API Response + Usage Data
-    Routstr->>DB: Deduct Actual Cost
-    Routstr-->>Client: API Response
-```
+---
 
-## Quick Links
+## 🔌 API Reference
 
-<div class="grid cards" markdown>
+- **[Overview](api/overview.md)**: Base URL, headers, and standards.
+- **[Endpoints](api/endpoints.md)**: Full list of REST endpoints.
+- **[Authentication](api/authentication.md)**: Handling API keys and tokens.
+- **[Errors](api/errors.md)**: Status codes and debugging.
 
-- :rocket: **[Quick Start](getting-started/quickstart.md)**
+## 🛠️ Contributing
 
-    Get up and running with Docker in minutes
+- **[Architecture](contributing/architecture.md)**: System design.
+- **[Setup](contributing/setup.md)**: Development environment.
+- **[Testing](contributing/testing.md)**: Running tests.
 
-- :gear: **[Configuration](getting-started/configuration.md)**
+---
 
-    Learn about environment variables and settings
-
-- :book: **[User Guide](user-guide/introduction.md)**
-
-    Comprehensive guide for using Routstr
-
-- :hammer: **[Contributing](contributing/setup.md)**
-
-    Help improve Routstr Core
-
-</div>
-
-## Use Cases
-
-- **AI Application Developers** - Add Bitcoin payments to your AI apps without managing infrastructure
-- **API Resellers** - Resell API access with custom pricing and profit margins
-- **Privacy-Focused Users** - Access AI models without revealing personal information
-- **Micropayment Experiments** - Test new business models with instant, small payments
-
-## Getting Help
-
-- 📖 Browse the [User Guide](user-guide/introduction.md) for detailed usage instructions
-- 🐛 Report issues on [GitHub](https://github.com/routstr/routstr-core/issues)
-- 💬 Join the community discussions
-- 🔧 Check the [API Reference](api/overview.md) for technical details
-
-## License
-
-Routstr Core is open source software licensed under the GPLv3. See the [LICENSE](https://github.com/routstr/routstr-core/blob/main/LICENSE) file for details.
+*Powered by [Cashu](https://cashu.space) and [Nostr](https://nostr.com).*

docs/overview.md

@@ -0,0 +1,76 @@
+# Overview
+
+Routstr is a decentralized protocol for **permissionless, private, and censorship-resistant AI inference**. It creates an open marketplace where anyone can sell llm-tokens and anyone can buy them using privacy-preserving micropayments.
+
+By combining **Nostr** (for censorship-resistant discovery and communication) and **Cashu** (for private, instant Bitcoin eCash payments), Routstr effectively removes the "middleman" from the AI ecosystem.
+
+## How it Works
+
+The network consists of independent **Providers** (Sellers) and **Clients** (Buyers). There is no central server, no login, and no credit card required.
+
+1. **Discovery (Nostr)**: Providers announce their availability, models (e.g., `gpt-4o`, `deepseek-r1`), and prices on the Nostr network.
+2. **Payment (Cashu)**: Clients pay providers directly using Bitcoin eCash (Cashu tokens). These payments are untraceable and settle instantly.
+3. **Inference (Proxy)**: The Provider acts as a gateway (or runs local hardware), executing the AI model and returning the result to the Client.
+
+## Who is this for?
+
+The documentation is split into two paths depending on your goal:
+
+### 🐣 I want to BUILD on Routstr (Client)
+
+You are a developer building an AI agent, a chat app, or a script, and you want access to AI models without API keys, subscriptions, or KYC.
+
+* **No Accounts**: Just get a wallet.
+* **Privacy**: Your requests are mixed with thousands of others; providers can't profile you.
+* **Choice**: Switch between hundreds of providers instantly for the best price/performance.
+
+👉 **[Go to Client Guide](client/introduction.md)**
+
+### 🦁 I want to RUN a Node (Provider)
+
+You have API credits (OpenAI, Anthropic, etc.) or GPU capacity and want to earn Bitcoin by selling AI access to the network.
+
+* **Monetize API Keys**: Connect your OpenAI/Anthropic/OpenRouter accounts and earn sats on every request.
+* **Monetize Hardware**: Run local models (via vLLM, Ollama) and sell access.
+* **Permissionless**: No approval needed. Start the container, configure via dashboard, start earning.
+
+!!! note "Coming Soon"
+    Future versions will support node-to-node routing—run a gateway without needing your own AI provider credentials.
+
+👉 **[Go to Provider Guide](provider/quickstart.md)**
+
+---
+
+## Architecture
+
+Routstr is built on a modular stack defined by the [Routstr Improvement Protocols (RIPs)](https://github.com/routstr/rips).
+
+```mermaid
+flowchart LR
+    subgraph Client
+        A[App / Agent]
+    end
+
+    subgraph Provider
+        B[Routstr Node<br/>Proxy + Auth + Billing]
+    end
+
+    subgraph Upstream
+        C[OpenAI / Anthropic<br/>vLLM / Ollama / ...]
+    end
+
+    A -- "Request +<br/>Cashu Token" --> B
+    B -- "Forward<br/>Request" --> C
+    C -- "Response +<br/>Usage" --> B
+    B -- "Response +<br/>Refund Token" --> A
+```
+
+## Why Routstr?
+
+| Feature | Closed AI | Routstr |
+| :--- | :--- | :--- |
+| **Access** | Account, KYC, Credit Card | Permissionless, Bitcoin-native |
+| **Privacy** | Full Logging & Tracking | Blinded Payments, Ephemeral Sessions |
+| **Resilience** | Single Point of Failure | Decentralized Network |
+| **Pricing** | Fixed, Monopolistic | Dynamic, Market-driven |
+| **Global** | Geofenced | Borderless (Tor/I2P supported) |

docs/provider/advanced-pricing.md

@@ -0,0 +1,114 @@
+# Advanced Pricing
+
+Advanced pricing strategies for fine-tuned control over your revenue model.
+
+---
+
+## Default Behavior
+
+By default, Routstr:
+
+1. **Fetches costs** from your upstream provider
+2. **Applies markup** using your fee settings
+3. **Converts to sats** using real-time BTC price
+
+**Formula**: `Price = Upstream Cost × Exchange Fee × Upstream Fee`
+
+---
+
+## Strategy 1: Fixed Per-Request
+
+Charge a flat fee regardless of model or tokens used.
+
+**Configure in Dashboard** → **Settings** → **Pricing**:
+
+- Enable **Fixed Pricing**
+- Set **Fixed Cost Per Request** (in sats)
+
+**Use cases**:
+
+- Internal tools with predictable usage
+- Simple "pay once, get response" APIs
+- Subscription-like tiers
+
+---
+
+## Strategy 2: Fixed Per-Token
+
+Override dynamic pricing with global per-token rates.
+
+**Configure in Dashboard** → **Settings** → **Pricing**:
+
+| Setting | Description |
+|---------|-------------|
+| **Fixed Per 1K Input** | Sats per 1,000 prompt tokens |
+| **Fixed Per 1K Output** | Sats per 1,000 completion tokens |
+
+When set to non-zero values, these override model-specific pricing for all models.
+
+---
+
+## Strategy 3: Per-Model Custom Pricing
+
+Set specific prices for individual models, overriding both upstream cost and global fees.
+
+**Configure in Dashboard** → **Models**:
+
+1. Click on a model (e.g., `gpt-4`)
+2. Enter **Prompt Price** and **Completion Price** (USD per 1M tokens)
+3. Save
+
+**Example**: OpenAI charges $30/1M for GPT-4. Set your price to $35/1M to lock in a margin regardless of fee settings.
+
+---
+
+## Minimum Charge
+
+Prevent dust transactions and spam:
+
+| Setting | Description | Default |
+|---------|-------------|---------|
+| **Min Request Cost** | Minimum charge in msats | 1000 (1 sat) |
+
+If a request's calculated cost falls below this (e.g., very short prompts), the client pays the minimum.
+
+---
+
+## Combining Strategies
+
+Strategies apply in order of specificity:
+
+1. **Per-model override** (highest priority)
+2. **Fixed per-token rates**
+3. **Dynamic pricing with fees** (default)
+4. **Fixed per-request** (overrides all above if enabled)
+
+**Example setup**:
+
+- Dynamic pricing as default (10% markup)
+- GPT-4 locked at $35/1M (premium model)
+- Claude Haiku at 5 sats/1K tokens (budget option)
+- Minimum 1 sat per request
+
+---
+
+## Pricing for Profit
+
+### High-Volume Strategy
+
+Lower margins, more clients:
+
+- Exchange Fee: 1.002 (0.2%)
+- Upstream Fee: 1.05 (5%)
+
+### Premium Strategy
+
+Higher margins, fewer clients:
+
+- Exchange Fee: 1.01 (1%)
+- Upstream Fee: 1.25 (25%)
+
+### Mixed Strategy
+
+- Cheap models (GLM-4.7-Flash, Seed-1.6): Low margin to attract volume
+- Premium models (GPT-5-Pro, Claude-Opus): High margin for profit

docs/provider/configuration.md

@@ -0,0 +1,122 @@
+# Configuration
+
+Routstr is configured primarily through the **Admin Dashboard**. All settings persist in the database and take effect immediately—no restarts required.
+
+For automated deployments, you can optionally pre-configure settings via environment variables.
+
+---
+
+## Admin Dashboard (Primary)
+
+Access the dashboard at `/admin/` on your node.
+
+### Upstream Providers
+
+Connect to your AI provider(s):
+
+| Setting | Description |
+|---------|-------------|
+| **Upstream URL** | API endpoint (e.g., `https://api.openai.com/v1`) |
+| **API Key** | Your provider's API key |
+
+### Node Identity
+
+How your node appears to clients:
+
+| Setting | Description |
+|---------|-------------|
+| **Name** | Display name (e.g., "Fast GPT-4 Node") |
+| **Description** | Brief description of your service |
+
+### Pricing
+
+Control your profit margins:
+
+| Setting | Description | Default |
+|---------|-------------|---------|
+| **Fixed Pricing** | Charge flat rate per request vs. per-token | Off |
+| **Exchange Fee** | Buffer for BTC volatility | 1.005 (0.5%) |
+| **Upstream Fee** | Your profit markup | 1.10 (10%) |
+
+See [Pricing](pricing.md) for detailed strategies.
+
+### Cashu Mints
+
+Which mints to accept payments from:
+
+| Setting | Description |
+|---------|-------------|
+| **Mints** | List of trusted Cashu mint URLs |
+
+### Lightning Withdrawals
+
+Automatic profit withdrawal:
+
+| Setting | Description |
+|---------|-------------|
+| **Lightning Address** | Your LN address for withdrawals |
+
+### Security
+
+| Setting | Description |
+|---------|-------------|
+| **Admin Password** | Password for dashboard access |
+
+### Nostr Discovery
+
+Announce your node on the network:
+
+| Setting | Description |
+|---------|-------------|
+| **Npub** | Your Nostr public key |
+| **Nsec** | Your Nostr private key (for signing) |
+| **Relays** | Relays to publish announcements |
+
+See [Discovery](discovery.md) for details.
+
+---
+
+## Environment Variables (Optional)
+
+Use environment variables for:
+
+- **Automated deployments** (CI/CD, infrastructure-as-code)
+- **Secrets management** (external secret stores)
+- **Initial bootstrap** (set once, manage via dashboard later)
+
+### All Variables
+
+| Variable | Description | Default |
+|----------|-------------|---------|
+| `UPSTREAM_BASE_URL` | Upstream API endpoint | — |
+| `UPSTREAM_API_KEY` | Upstream API key | — |
+| `ADMIN_PASSWORD` | Dashboard password | (none) |
+| `DATABASE_URL` | Database connection string | `sqlite+aiosqlite:///keys.db` |
+| `NAME` | Node display name | `ARoutstrNode` |
+| `DESCRIPTION` | Node description | `A Routstr Node` |
+| `NPUB` | Nostr public key (bech32) | — |
+| `NSEC` | Nostr private key | — |
+| `CASHU_MINTS` | Comma-separated mint URLs | `https://mint.minibits.cash/Bitcoin` |
+| `RECEIVE_LN_ADDRESS` | Lightning address for withdrawals | — |
+| `TOR_PROXY_URL` | SOCKS5 proxy for Tor | `socks5://127.0.0.1:9050` |
+| `CORS_ORIGINS` | Allowed CORS origins | `*` |
+| `RELAYS` | Nostr relays (comma-separated) | (default set) |
+
+### Priority
+
+Environment variables are read on startup. Dashboard settings override them and persist in the database. Once you change a setting in the dashboard, the env var is ignored for that setting.
+
+---
+
+## Models
+
+Manage which AI models you offer:
+
+1. Go to **Models** in the dashboard
+2. Models are auto-discovered from your upstream
+3. For each model, you can:
+   - **Enable/Disable** — hide expensive models you don't want to serve
+   - **Override pricing** — set custom per-token rates
+   - **Create aliases** — friendly names for models
+
+See [Pricing](pricing.md) for per-model pricing strategies.

docs/provider/dashboard.md

@@ -0,0 +1,208 @@
+# Admin Dashboard
+
+The Admin Dashboard is your command center for managing your Routstr provider node. Configure providers, monitor earnings, manage models, and withdraw profits—all from a web interface.
+
+**URL**: `http://your-node:8000/admin/`
+
+---
+
+## Overview Tab
+
+The main dashboard view shows your node's financial status at a glance.
+
+### Wallet Summary
+
+| Metric | Description |
+|--------|-------------|
+| **Total Wallet** | All Bitcoin currently held by your node |
+| **User Balances** | Funds belonging to active client sessions |
+| **Your Balance** | Your profit: `Total - User Balances` |
+
+### Mint Status
+
+Shows connected Cashu mints and their balances. Each mint displays:
+
+- Connection status
+- Balance in sats/msats
+- Unit type
+
+<!-- TODO: Screenshot of Overview tab -->
+
+---
+
+## Sessions Tab
+
+View and manage active client sessions (API keys).
+
+### Session List
+
+| Column | Description |
+|--------|-------------|
+| **Hashed Key** | Privacy-preserving identifier (not the actual key) |
+| **Balance** | Remaining funds in the session |
+| **Spent** | Total amount spent by this session |
+| **Requests** | Number of API calls made |
+| **Created** | When the session was created |
+| **Expires** | Auto-expiry time (if set) |
+
+### Actions
+
+- **View Details** — See full session history
+- **Revoke** — Terminate a session (remaining balance returns to your wallet)
+
+<!-- TODO: Screenshot of Sessions tab -->
+
+---
+
+## Models Tab
+
+Manage which AI models you offer to clients.
+
+### Model List
+
+Shows all models available from your upstream provider(s):
+
+| Column | Description |
+|--------|-------------|
+| **Model ID** | The model identifier (e.g., `gpt-4o`) |
+| **Enabled** | Whether clients can use this model |
+| **Input Price** | Cost per 1M input tokens (USD) |
+| **Output Price** | Cost per 1M output tokens (USD) |
+| **Custom** | Whether pricing is overridden |
+
+### Actions
+
+- **Import Models** — Fetch latest model list from upstream
+- **Enable/Disable** — Toggle model availability
+- **Edit Pricing** — Override default pricing for a model
+- **Create Alias** — Map a friendly name to a model
+
+### Editing a Model
+
+Click on any model to configure:
+
+| Field | Description |
+|-------|-------------|
+| **Enabled** | Show this model to clients |
+| **Prompt Price** | Custom price per 1M input tokens (USD) |
+| **Completion Price** | Custom price per 1M output tokens (USD) |
+| **Alias** | Alternative name for this model |
+
+<!-- TODO: Screenshot of Models tab -->
+<!-- TODO: Screenshot of Model edit modal -->
+
+---
+
+## Settings Tab
+
+Configure all node settings. Changes take effect immediately.
+
+### Upstream
+
+Connect to your AI provider:
+
+| Field | Description |
+|-------|-------------|
+| **Base URL** | API endpoint (e.g., `https://api.openai.com/v1`) |
+| **API Key** | Your provider's secret key |
+
+<!-- TODO: Screenshot of Upstream settings -->
+
+### Node Identity
+
+| Field | Description |
+|-------|-------------|
+| **Name** | Public display name |
+| **Description** | Brief description of your service |
+| **Npub** | Nostr public key for discovery |
+
+### Pricing
+
+| Field | Description |
+|-------|-------------|
+| **Fixed Pricing** | Toggle flat-rate vs. per-token pricing |
+| **Fixed Cost** | Sats per request (when fixed pricing enabled) |
+| **Exchange Fee** | Multiplier for BTC volatility buffer |
+| **Upstream Fee** | Your profit margin multiplier |
+
+**Example**: With Exchange Fee `1.005` and Upstream Fee `1.10`:
+
+- Upstream cost: $30/1M tokens
+- Your price: $30 × 1.005 × 1.10 = $33.17/1M tokens
+
+### Cashu Mints
+
+Manage which mints you accept payments from:
+
+- **Add Mint** — Enter a mint URL
+- **Remove Mint** — Stop accepting from a mint
+- **Test Connection** — Verify mint is reachable
+
+<!-- TODO: Screenshot of Mints settings -->
+
+### Lightning
+
+| Field | Description |
+|-------|-------------|
+| **Lightning Address** | Your LN address for automatic withdrawals |
+
+### Nostr Discovery
+
+| Field | Description |
+|-------|-------------|
+| **Nsec** | Private key for signing announcements |
+| **Relays** | Where to publish your node advertisement |
+
+### Security
+
+| Field | Description |
+|-------|-------------|
+| **Admin Password** | Password for dashboard access |
+
+!!! warning "Set a Password"
+    The dashboard has no password by default. Always set one for production nodes.
+
+<!-- TODO: Screenshot of Security settings -->
+
+---
+
+## Withdraw Tab
+
+Withdraw your profits to a Lightning wallet.
+
+### Steps
+
+1. **Select Mint** — Choose which mint to withdraw from
+2. **Enter Amount** — How many sats to withdraw
+3. **Generate Token** — Creates a Cashu token
+4. **Redeem** — Paste the token into your Cashu wallet and melt to Lightning
+
+<!-- TODO: Screenshot of Withdraw tab -->
+
+### Alternative: Lightning Address
+
+If you've configured a Lightning Address in Settings, profits can be automatically swept to your wallet (coming soon).
+
+---
+
+## Logs Tab
+
+View node logs for debugging without SSH access.
+
+### Features
+
+- **Filter by Level** — Error, Warning, Info, Debug
+- **Search** — Find specific entries
+- **Time Range** — View logs from specific periods
+- **Auto-refresh** — Watch logs in real-time
+
+### Common Log Entries
+
+| Entry | Meaning |
+|-------|---------|
+| `Upstream request failed` | Problem connecting to your AI provider |
+| `Invalid token` | Client sent an invalid Cashu token |
+| `Session expired` | API key reached its time limit |
+| `Insufficient balance` | Client ran out of funds mid-request |
+
+<!-- TODO: Screenshot of Logs tab -->

docs/provider/deployment.md

@@ -0,0 +1,203 @@
+# Deployment
+
+Production deployment guide for Routstr Provider nodes.
+
+## Docker Compose (Recommended)
+
+For production, use Docker Compose with persistent storage and optional Tor support.
+
+### Unified Setup (All-in-one)
+To build and run the node with the UI integrated in a single container using the multi-stage build:
+
+```bash
+docker build -f Dockerfile.full -t routstr-full .
+docker run -d -p 8000:8000 --env-file .env routstr-full
+```
+
+### Advanced Setup (Separated UI & Node)
+Use the included `compose.yml` for a more flexible setup that separates the UI build process from the node execution. This is useful for development or when you want to manage Tor as a separate service.
+
+```bash
+docker compose up -d
+```
+
+This will:
+1.  **Build the UI**: Compiles the frontend and copies it to a shared volume.
+2.  **Start Routstr**: Runs the Python node, mounting the built UI.
+3.  **Start Tor**: Provides anonymous access via a `.onion` address.
+
+---
+
+## With Tor (Anonymous Access)
+
+Add Tor to serve your node as a hidden service—no port forwarding needed.
+
+```yaml
+services:
+  routstr:
+    image: ghcr.io/routstr/proxy:latest
+    container_name: routstr
+    restart: unless-stopped
+    ports:
+      - "8000:8000"
+    volumes:
+      - ./data:/app/data
+      - ./logs:/app/logs
+    environment:
+      - TOR_PROXY_URL=socks5://tor:9050
+    depends_on:
+      - tor
+
+  tor:
+    image: ghcr.io/hundehausen/tor-hidden-service:latest
+    container_name: tor
+    restart: unless-stopped
+    volumes:
+      - ./tor-data:/var/lib/tor
+    environment:
+      - HS_ROUTER=routstr:8000:80
+```
+
+After starting, find your `.onion` address:
+
+```bash
+docker exec tor cat /var/lib/tor/hidden_service/hostname
+```
+
+See [Tor Support](tor.md) for details.
+
+---
+
+## Pre-Configuration (Optional)
+
+While everything can be configured via the dashboard, you can pre-configure settings with environment variables for automated deployments.
+
+### Using Environment Variables
+
+```yaml
+services:
+  routstr:
+    image: ghcr.io/routstr/proxy:latest
+    environment:
+      # Pre-configure upstream (optional)
+      - UPSTREAM_BASE_URL=https://api.openai.com/v1
+      - UPSTREAM_API_KEY=sk-proj-...
+      
+      # Secure the dashboard (recommended)
+      - ADMIN_PASSWORD=your-secure-password
+      
+      # Node identity
+      - NAME=My Provider Node
+      - DESCRIPTION=Fast GPT-4 access via Lightning
+      
+      # Lightning withdrawals
+      - RECEIVE_LN_ADDRESS=me@walletofsatoshi.com
+    volumes:
+      - ./data:/app/data
+```
+
+### Using an .env File
+
+```yaml
+services:
+  routstr:
+    image: ghcr.io/routstr/proxy:latest
+    env_file:
+      - .env
+    volumes:
+      - ./data:/app/data
+```
+
+Example `.env`:
+
+```bash
+UPSTREAM_BASE_URL=https://api.openai.com/v1
+UPSTREAM_API_KEY=sk-proj-...
+ADMIN_PASSWORD=change-me
+NAME=My Provider Node
+RECEIVE_LN_ADDRESS=me@walletofsatoshi.com
+```
+
+See [Configuration](configuration.md) for all available options.
+
+---
+
+## Persistence
+
+Routstr stores all data in `/app/data`:
+
+| Path | Contents |
+|------|----------|
+| `keys.db` | SQLite database (settings, API keys, sessions) |
+| `.wallet/` | Cashu wallet data (your Bitcoin!) |
+
+!!! warning "Back Up Your Data"
+    The `./data` volume contains your wallet. Losing it means losing funds. Back up regularly.
+
+---
+
+## Reverse Proxy (Optional)
+
+For custom domains and SSL, use a reverse proxy like Caddy or nginx.
+
+### Caddy Example
+
+```
+api.yournode.com {
+    reverse_proxy localhost:8000
+}
+```
+
+### nginx Example
+
+```nginx
+server {
+    listen 443 ssl;
+    server_name api.yournode.com;
+    
+    ssl_certificate /path/to/cert.pem;
+    ssl_certificate_key /path/to/key.pem;
+    
+    location / {
+        proxy_pass http://localhost:8000;
+        proxy_http_version 1.1;
+        proxy_set_header Upgrade $http_upgrade;
+        proxy_set_header Connection "upgrade";
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+    }
+}
+```
+
+---
+
+## Updates
+
+Pull the latest image and restart:
+
+```bash
+docker compose pull
+docker compose up -d
+```
+
+---
+
+## Building from Source
+
+### Unified Image (UI + Node)
+The easiest way to build everything from source into a single production-ready image:
+
+```bash
+docker build -f Dockerfile.full -t routstr-full .
+```
+
+### Individual Components
+If you prefer building them separately or using Docker Compose:
+
+```bash
+# Build using compose
+docker compose build
+
+# Or build the node only (requires manual UI build first)
+docker build -t routstr-node .
+```

docs/provider/discovery.md

@@ -0,0 +1,95 @@
+# Discovery
+
+Routstr uses **Nostr** as a decentralized directory for service discovery. Your node announces its presence, models, and pricing on Nostr relays, allowing clients to find you without a central server.
+
+---
+
+## How It Works
+
+1. **Provider Advertisement (Kind 38421)**: Your node periodically publishes an event with its URL, models, and pricing
+2. **Client Discovery**: Clients query relays for these events to find suitable providers
+
+---
+
+## Configuration
+
+Configure discovery in **Dashboard** → **Settings** → **Nostr**.
+
+### Required Settings
+
+| Field | Description |
+|-------|-------------|
+| **Npub** | Your node's public identity (clients use this to verify your node) |
+| **Nsec** | Your node's private key (used to sign advertisements) |
+| **Relays** | Where to publish your announcements |
+
+### Default Relays
+
+If not configured, Routstr publishes to:
+
+- `wss://relay.damus.io`
+- `wss://relay.nostr.band`
+- `wss://nos.lol`
+
+---
+
+## Advertisement Format
+
+Your node publishes events like:
+
+```json
+{
+  "kind": 38421,
+  "content": {
+    "name": "My Routstr Node",
+    "description": "Fast GPT-4 access via Lightning",
+    "endpoints": {
+      "http": "https://api.mynode.com",
+      "onion": "http://xyz...onion"
+    },
+    "models": ["gpt-4", "claude-3-opus"],
+    "pricing": { ... }
+  },
+  "tags": [
+    ["d", "routstr-provider"],
+    ["g", "US"]
+  ]
+}
+```
+
+---
+
+## Tor Integration
+
+If you're running with Tor (see [Tor Support](tor.md)), your `.onion` address is automatically included in announcements. This allows clients to connect anonymously.
+
+---
+
+## Verify Your Announcements
+
+Check if your node is broadcasting:
+
+1. Copy your `Npub`
+2. Search on [Nostr.band](https://nostr.band) or [Primal](https://primal.net)
+3. Look for Kind 38421 events
+
+---
+
+## Generating Keys
+
+If you don't have a Nostr identity:
+
+1. Use any Nostr client (e.g., [Primal](https://primal.net), [Damus](https://damus.io))
+2. Create an account
+3. Export your keys (npub and nsec)
+4. Enter them in the dashboard
+
+Or generate keys programmatically:
+
+```python
+from nostr_sdk import Keys
+
+keys = Keys.generate()
+print(f"npub: {keys.public_key().to_bech32()}")
+print(f"nsec: {keys.secret_key().to_bech32()}")
+```

docs/provider/pricing.md

@@ -0,0 +1,91 @@
+# Pricing
+
+Routstr's pricing engine lets you act as a retailer of AI compute. You pay upstream providers (OpenAI, Anthropic, etc.) at their rates and sell to clients with your markup.
+
+---
+
+## Pricing Strategies
+
+Configure these in **Dashboard** → **Settings** → **Pricing**.
+
+### Dynamic Pricing (Default)
+
+Passes through upstream costs plus your percentage markup.
+
+**Formula**: `Client Price = Upstream Cost × Exchange Fee × Upstream Fee`
+
+| Setting | Description | Default |
+|---------|-------------|---------|
+| **Exchange Fee** | Buffer for BTC price volatility | 1.005 (0.5%) |
+| **Upstream Fee** | Your profit margin | 1.10 (10%) |
+
+**Example**: GPT-4 costs $30/1M tokens from OpenAI. With default settings:
+
+- Price: $30 × 1.005 × 1.10 = $33.17/1M tokens
+- At $60k BTC: ~55,000 sats/1M tokens
+
+### Fixed Pricing
+
+Charge a flat rate per request, regardless of model or token count.
+
+| Setting | Description |
+|---------|-------------|
+| **Fixed Pricing** | Enable flat-rate mode |
+| **Fixed Cost** | Sats per request |
+
+**Best for**: Simple proxies, internal tools, or subscription-like access.
+
+---
+
+## Per-Model Pricing
+
+Override pricing for specific models in **Dashboard** → **Models**.
+
+1. Click on a model
+2. Enter custom **Prompt Price** and **Completion Price** (USD per 1M tokens)
+3. Save
+
+This overrides both the upstream cost and your global markup for that model.
+
+**Example**: Lock GPT-4 at $35/1M tokens regardless of OpenAI's actual rate or your fee settings.
+
+---
+
+## Token-Based Overrides
+
+Set global fixed rates per token (overrides dynamic pricing for all models):
+
+| Setting | Description |
+|---------|-------------|
+| **Fixed Per 1K Input** | Sats per 1,000 prompt tokens |
+| **Fixed Per 1K Output** | Sats per 1,000 completion tokens |
+
+---
+
+## Minimum Charge
+
+Prevent spam with a minimum cost per request:
+
+| Setting | Description | Default |
+|---------|-------------|---------|
+| **Min Request Cost** | Minimum charge in msats | 1000 (1 sat) |
+
+If a request's calculated cost is lower than this, the client pays the minimum instead.
+
+---
+
+## Cost Tracking
+
+Routstr tracks balances in **millisats (msats)** for precision with cheap models.
+
+- 1 sat = 1,000 msats
+- API responses include cost in msats
+- Lightning withdrawals round down to whole sats
+
+### Client Verification (RIP-05)
+
+Clients can verify charges:
+
+1. Fetch `/v1/models` for your advertised rates
+2. Calculate expected cost from token counts
+3. Compare to `x-routstr-cost` response header

docs/provider/quickstart.md

@@ -0,0 +1,111 @@
+# Quick Start
+
+Start earning Bitcoin by selling AI access in under 5 minutes.
+
+## What You'll Build
+
+A **Routstr Provider Node** acts as a gateway that:
+
+1. **Connects** to upstream AI providers (OpenAI, Anthropic, OpenRouter, etc.)
+2. **Accepts** Bitcoin payments via Cashu eCash
+3. **Serves** AI requests to clients on the network
+
+You bring the API keys, Routstr handles the billing, payments, and client management.
+
+!!! tip "Future: Node-to-Node Routing"
+    In future versions, you'll be able to run a node that connects to other Routstr nodes—eliminating the need to configure upstream providers yourself. For now, you'll need your own API credentials.
+
+---
+
+## Prerequisites
+
+- [Docker](https://docs.docker.com/get-docker/) installed
+- API credentials from at least one AI provider (OpenAI, Anthropic, OpenRouter, etc.)
+
+---
+
+## 1. Start the Node
+
+You can run the pre-built image directly:
+
+```bash
+docker run -d \
+  --name routstr \
+  -p 8000:8000 \
+  -v routstr-data:/app/data \
+  ghcr.io/routstr/proxy:latest
+```
+
+### Build from Source (Recommended)
+If you want to build the node and UI yourself from source, use the unified Dockerfile:
+
+```bash
+git clone https://github.com/routstr/routstr-core.git
+cd routstr-core
+docker build -f Dockerfile.full -t routstr-local .
+docker run -d -p 8000:8000 --name routstr routstr-local
+```
+
+Verify it's running:
+
+```bash
+curl http://localhost:8000/v1/info
+```
+
+---
+
+## 2. Configure via Dashboard
+
+Open the **Admin Dashboard** at [http://localhost:8000/admin/](http://localhost:8000/admin/).
+
+!!! note "Default Access"
+    The dashboard has no password by default. Set one immediately in Settings for production use.
+
+### Connect Your AI Providers
+
+1. Navigate to **Settings** → **Upstream**
+2. Enter your upstream URL (e.g., `https://api.openai.com/v1`)
+3. Enter your API key
+4. Save
+
+### Set Your Profit Margin
+
+1. Go to **Settings** → **Pricing**
+2. Configure your markup (default is 10%)
+3. Optionally set a fixed price per request instead
+
+### Secure the Dashboard
+
+1. Go to **Settings** → **Admin**
+2. Set a strong password
+3. Save and re-login
+
+---
+
+## 3. Start Earning
+
+Once configured, your node is live. Clients pay you in Bitcoin (via Cashu tokens) for every AI request.
+
+### Monitor Your Earnings
+
+The dashboard shows:
+
+- **Total Wallet**: All Bitcoin held by your node
+- **User Balances**: Funds belonging to active client sessions
+- **Your Balance**: Your profit (`Total - User Balances`)
+
+### Withdraw Profits
+
+1. Go to **Withdraw** in the dashboard
+2. Select amount and mint
+3. Generate a Cashu token
+4. Redeem to your Lightning wallet
+
+---
+
+## Next Steps
+
+- **[Deployment](deployment.md)**: Production setup with Docker Compose and Tor
+- **[Dashboard Guide](dashboard.md)**: Full reference for all dashboard features
+- **[Pricing](pricing.md)**: Configure pricing strategies and per-model overrides
+- **[Discovery](discovery.md)**: Announce your node on Nostr for clients to find you

docs/provider/tor.md

@@ -0,0 +1,50 @@
+# Tor Support
+
+Running Routstr as a **Tor Hidden Service** allows you to offer API access anonymously and bypass NAT/firewalls without port forwarding.
+
+## Automatic Setup (Docker)
+
+The standard `compose.yml` includes a Tor container pre-configured to serve your node.
+
+1.  **Start the stack**: `docker compose up -d`
+2.  **Wait**: Tor takes about 30 seconds to generate keys and bootstrap.
+3.  **Find your address**:
+    ```bash
+    docker exec tor cat /var/lib/tor/hidden_service/hostname
+    ```
+    Output: `v2xyz...longaddress.onion`
+
+Routstr will automatically detect this address (via the `discover_onion_url_from_tor` logic) and include it in:
+- The `/v1/info` endpoint.
+- Nostr announcements (RIP-02).
+
+## Manual Setup
+
+If you are running outside Docker or managing Tor yourself:
+
+1.  **Install Tor**: `sudo apt install tor`
+2.  **Edit `torrc`**:
+    ```
+    HiddenServiceDir /var/lib/tor/routstr/
+    HiddenServicePort 80 127.0.0.1:8000
+    ```
+3.  **Restart Tor**: `sudo systemctl restart tor`
+4.  **Get Address**: `sudo cat /var/lib/tor/routstr/hostname`
+5.  **Configure Routstr**:
+    Set `ONION_URL=http://youraddress.onion` in your `.env` file so the node knows its own address.
+
+## Client Usage
+
+Clients connecting to your `.onion` address must route traffic through SOCKS5.
+
+**Python Example:**
+```python
+import httpx
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="http://youraddress.onion/v1",
+    api_key="sk-...",
+    http_client=httpx.Client(proxy="socks5://127.0.0.1:9050")
+)
+```

docs/user-guide/admin-dashboard.md

@@ -1,216 +0,0 @@
-# Admin Dashboard
-
-The Routstr admin dashboard is a modern web interface for managing your node, monitoring wallet balances, configuring AI models and providers, and handling Bitcoin Lightning payments through Cashu eCash.
-
-## Accessing the Dashboard
-
-### Authentication
-
-The dashboard is protected by password authentication:
-
-1. Navigate to `/admin/` in your browser
-2. Enter the admin password
-3. Optional: Configure custom base URL if not pre-configured
-4. Click "Login"
-
-The interface supports both environment-configured URLs and manual URL entry for deployment flexibility.
-
-## Dashboard Overview
-
-The main dashboard consists of four primary sections accessible through a collapsible sidebar:
-
-- **Dashboard** - Wallet balance monitoring and fund management
-- **Models** - AI model management and testing
-- **Providers** - Upstream provider configuration
-- **Settings** - Node configuration and admin preferences
-
-### Navigation
-
-## Dashboard Page
-
-### Wallet Balance Management
-
-#### Balance Display Options
-
-Switch between display units using the toggle buttons:
-
-- **msat** - Millisatoshis (highest precision)
-- **sat** - Satoshis (standard Bitcoin unit)
-- **usd** - US Dollar equivalent (when exchange rate available)
-
-#### Balance Overview
-
-The dashboard displays three key metrics:
-
-- **Your Balance (Total)** - Available funds for node operator
-- **Total Wallet** - Combined balance across all Cashu mints
-- **User Balance** - Funds held for API key holders
-
-#### Detailed Balance Breakdown
-
-View balances by mint with the following information:
-
-| Column      | Description                           |
-| ----------- | ------------------------------------- |
-| Mint / Unit | Cashu mint URL and currency unit      |
-| Wallet      | Total funds in this mint              |
-| Users       | Funds belonging to API key holders    |
-| Owner       | Your available funds (Wallet - Users) |
-
-### Temporary Balances
-
-Monitor API key activity with:
-
-- **Summary Cards** - Total balance, total spent, total requests
-- **Search Functionality** - Filter by key hash or refund address
-- **Detailed Table** - Individual key balances with expiry times
-- **Auto-refresh** - Updates every 60 seconds
-
-### Fund Management
-
-#### Withdrawing Funds
-
-To withdraw your available balance:
-
-1. Click the **Withdraw** button
-2. Select which mint to withdraw from
-3. Specify the amount (or withdraw full balance)
-4. Click **Generate Token**
-5. Copy the generated eCash token
-6. Import the token into your Cashu wallet
-
-#### Real-time Updates
-
-- Balances refresh automatically every 30 seconds
-- Manual refresh option available
-- Live Bitcoin/USD exchange rate integration
-- Error handling for mint connectivity issues
-
-## Models Management Page
-
-### Model Organization
-
-Models are organized by provider groups with tabs:
-
-- **All Models** - Combined view of all available models
-- **Provider-specific tabs** - Individual providers (OpenRouter, Azure, etc.)
-- Badge indicators showing active/total model counts
-
-### Model Management Features
-
-#### Individual Model Operations
-
-For each model you can:
-
-- **Toggle Enable/Disable** - Control model availability
-- **View Details** - Context length, pricing, description
-- **Edit Configuration** - Model-specific settings
-- **Status Indicators** - Green badges for enabled, gray for disabled
-
-#### Bulk Operations
-
-- **Select All/Deselect All** - Quick selection controls
-- **Bulk Enable/Disable** - Mass model management
-- **Bulk Delete** - Remove model overrides
-- **Provider-level Actions** - Apply settings to all models in a provider
-
-#### Model Information Display
-
-- **Model Types** - Text, embedding, image, audio, multimodal indicators
-- **Pricing Information** - Per-million-token costs for input/output
-- **Context Length** - Maximum tokens supported
-- **API Key Status** - Whether credentials are configured
-- **Free Model Indicators** - No-cost models clearly marked
-
-## Providers Management Page
-
-### Upstream Provider Configuration
-
-Manage AI provider connections and credentials:
-
-#### Provider Types Supported
-
-- **OpenRouter** - Multi-model aggregator
-- **Azure OpenAI** - Microsoft's OpenAI service
-- **OpenAI** - Direct OpenAI integration
-- **Custom Providers** - Any OpenAI-compatible API
-
-#### Adding New Providers
-
-1. Click **Add Provider**
-2. Select **Provider Type** from dropdown
-3. Enter **Base URL** (auto-populated for known providers)
-4. Add **API Key** for authentication
-5. Set **API Version** (required for Azure)
-6. Toggle **Enabled** status
-7. Click **Create**
-
-#### Provider Management
-
-**Provider Cards Display:**
-
-- Provider type and status (Enabled/Disabled)
-- Base URL configuration
-- Action buttons (Models, Edit, Delete)
-
-**Available Actions:**
-
-- **Edit** - Modify provider configuration
-- **Delete** - Remove provider (with confirmation)
-- **View Models** - Expand model discovery interface
-- **Enable/Disable** - Toggle provider availability
-
-#### Model Discovery
-
-Each provider shows two types of models:
-
-**Provided Models Tab:**
-
-- Auto-discovered from provider's catalog
-- Read-only model information
-- Real-time availability updates
-
-**Custom Models Tab:**
-
-- Manually configured model overrides
-- Extend or override provider catalog
-- Individual enable/disable controls
-
-## Settings Page
-
-### Node Configuration
-
-Configure core node settings and preferences:
-
-#### Basic Information
-
-- **Node Name** - Identifier for your node
-- **Node Description** - Descriptive text for your service
-- **HTTP URL** - Public HTTP endpoint
-- **Onion URL** - Tor hidden service address
-
-#### Nostr Integration
-
-- **Public Key (npub)** - Your Nostr public identity
-- **Private Key (nsec)** - Nostr private key with show/hide toggle
-- **Nostr Relays** - Configure relays for provider announcements
-
-#### Cashu Mint Management
-
-- **Add Mint URLs** - Configure multiple Cashu mint endpoints
-- **Remove Mints** - Delete unused mint configurations
-- **Mint Validation** - Verify mint endpoint connectivity
-
-#### Settings Features
-
-- **Real-time Save** - Changes apply immediately
-- **Validation** - Form validation with error feedback
-- **Secure Fields** - Password masking with reveal toggles
-- **Reload Functionality** - Refresh configuration from server
-
-## Next Steps
-
-- [Payment Flow](payment-flow.md) - Understanding Bitcoin payment processing
-- [Using the API](using-api.md) - Making API requests to your node
-- [Models & Pricing](models-pricing.md) - Configuring model pricing and fees
-- [API Reference](../api/overview.md) - Complete API documentation

docs/user-guide/introduction.md

@@ -1,245 +0,0 @@
-# User Guide Introduction
-
-Welcome to the Routstr Core User Guide. This guide will help you understand how to use Routstr to access AI APIs with Bitcoin micropayments.
-
-## What You'll Learn
-
-- How the payment system works
-- Creating and managing API keys
-- Making API calls through Routstr
-- Using the admin dashboard
-- Managing your balance
-
-## Prerequisites
-
-Before starting, you'll need:
-
-1. **A Running Routstr Instance**
-   - Either your own deployment or access to a public node
-   - The base URL (e.g., `https://api.routstr.com/v1`)
-
-2. **A Cashu Wallet** (optional but recommended)
-   - [Nutstash](https://nutstash.app) - Web wallet
-   - [Minibits](https://www.minibits.cash) - Mobile wallet
-   - [Cashu.me](https://cashu.me) - Simple web wallet
-
-3. **An API Client**
-   - OpenAI Python/JavaScript SDK
-   - Any HTTP client (curl, Postman, etc.)
-   - Your application code
-
-## How Routstr Works
-
-### Traditional API Access
-
-```mermaid
-graph LR
-    A[Your App] --> B[OpenAI API]
-    B --> A
-```
-
-- Direct connection to provider
-- Monthly billing
-- Credit card required
-- Usage limits
-
-### With Routstr
-
-```mermaid
-graph LR
-    A[Your App] --> B[Routstr Proxy]
-    B --> C[OpenAI API]
-    C --> B
-    B --> A
-    D[Bitcoin/eCash] --> B
-```
-
-- Pay per request with Bitcoin
-- No credit card needed
-- Anonymous payments
-- Instant settlement
-
-## Key Concepts
-
-### eCash Tokens
-
-- Digital bearer tokens backed by Bitcoin
-- Can be sent like cash - whoever has the token owns it
-- Redeemable at Cashu mints for Bitcoin
-- Perfect for micropayments
-
-### API Keys
-
-- Created by depositing eCash tokens
-- Track your balance and usage
-- Can be topped up anytime
-- Optional expiry and refund address
-
-### Balance Management
-
-- Measured in millisatoshis (msats)
-- 1 Bitcoin = 100,000,000 sats = 100,000,000,000 msats
-- Deducted based on actual usage
-- Withdrawable as eCash tokens
-
-## Typical Workflow
-
-### 1. Get Bitcoin/eCash
-
-Options:
-
-- Buy Bitcoin and deposit to a Cashu mint
-- Receive eCash tokens from someone else
-- Use a testnet mint for testing
-
-### 2. Use Your eCash
-
-You have two options for using your eCash tokens with Routstr:
-
-#### Option A: Create a Persistent Wallet
-
-Create a wallet with an API key for multiple requests:
-
-```bash
-POST /v1/wallet/create
-{
-  "cashu_token": "cashuAeyJ0..."
-}
-```
-
-This returns an API key (`sk-...`) and your balance. The wallet persists between requests.
-
-#### Option B: Direct Token Usage
-
-Use your Cashu token directly as the API key:
-
-```python
-client = OpenAI(
-    api_key="cashuAeyJ0...",  # Your Cashu token directly
-    base_url="https://api.routstr.com/v1"
-)
-```
-
-Routstr automatically converts the token to access the associated wallet. Each request consumes from the token's balance.
-
-### 3. Make API Calls
-
-With either method:
-
-```python
-# Using persistent wallet API key
-client = OpenAI(
-    api_key="sk-...",
-    base_url="https://api.routstr.com/v1"
-)
-
-# Or using Cashu token directly
-client = OpenAI(
-    api_key="cashuAeyJ0...",
-    base_url="https://api.routstr.com/v1"
-)
-```
-
-### 4. Monitor Usage
-
-- Check balance: `GET /v1/wallet/balance`
-- View admin dashboard
-- Track costs per request
-
-### 5. Withdraw Funds
-
-When done, withdraw remaining balance as eCash through the admin interface.
-
-## Supported Endpoints
-
-Routstr supports all standard OpenAI endpoints:
-
-- ✅ `/v1/chat/completions` - Chat models
-- 🚧 `/v1/completions` - Text completion (Coming soon)
-- 🚧 `/v1/embeddings` - Text embeddings (Coming soon)
-- 🚧 `/v1/images/generations` - Image generation (Coming soon)
-- 🚧 `/v1/audio/transcriptions` - Audio to text (Coming soon)
-- 🚧 `/v1/audio/translations` - Audio translation (Coming soon)
-- ✅ `/v1/models` - List available models
-- ✅ Custom provider endpoints
-
-## Cost Structure
-
-### Pricing Models
-
-1. **Fixed Cost Per Request**
-   - Simple flat fee per API call
-   - Good for uniform usage
-
-2. **Token-Based Pricing**
-   - Pay per input/output token
-   - More accurate for varied usage
-
-3. **Model-Based Pricing**
-   - Different rates per model
-   - Reflects actual provider costs
-
-### Cost Calculation
-
-```
-Total Cost = Base Fee + (Input Tokens * Input Rate) + (Output Tokens * Output Rate)
-```
-
-Fees may include:
-
-- Exchange rate markup (BTC/USD conversion)
-- Provider margin
-- Node operator fee
-
-## Getting Support
-
-### Documentation
-
-- This user guide for general usage
-- [API Reference](../api/overview.md) for technical details
-- [Contributing Guide](../contributing/setup.md) for developers
-
-### Community
-
-- GitHub Issues for bugs and features
-- Nostr for decentralized discussion
-- Node operator contact info
-
-### Troubleshooting
-
-Common issues and solutions:
-
-- [Payment Flow](payment-flow.md) - Understanding the payment process
-- [Using the API](using-api.md) - API integration guide
-- [Admin Dashboard](admin-dashboard.md) - Managing your node
-
-## Security Considerations
-
-### API Key Security
-
-- Treat API keys like passwords
-- Never share or commit them
-- Rotate keys regularly
-- Use environment variables
-
-### Payment Security
-
-- eCash tokens are bearer instruments
-- Verify mint trustworthiness
-- Keep backups of tokens
-- Use small amounts for testing
-
-### Network Security
-
-- Always use HTTPS connections
-- Verify SSL certificates
-- Consider using Tor for privacy
-- Monitor for unusual activity
-
-## Next Steps
-
-Ready to start? Continue with:
-
-1. [Payment Flow](payment-flow.md) - Detailed payment process
-2. [Using the API](using-api.md) - Making your first calls
-3. [Admin Dashboard](admin-dashboard.md) - Managing your account

docs/user-guide/models-pricing.md

@@ -1,466 +0,0 @@
-# Models & Pricing
-
-Understanding how Routstr calculates costs is essential for managing your API usage efficiently. This guide explains the pricing models and how to configure them.
-
-## Pricing Models
-
-Routstr supports three pricing models:
-
-### 1. Fixed Pricing
-
-Simple per-request charging:
-
-```bash
-FIXED_PRICING=true
-FIXED_COST_PER_REQUEST=10  # 10 sats per request
-```
-
-**Best for:**
-
-- Uniform API usage
-- Simple applications
-- Predictable costs
-
-### 2. Token-Based Pricing
-
-Charge based on actual token usage:
-
-```bash
-FIXED_PRICING=false             # use model pricing
-FIXED_COST_PER_REQUEST=1        # optional base fee
-FIXED_PER_1K_INPUT_TOKENS=5     # optional override
-FIXED_PER_1K_OUTPUT_TOKENS=15   # optional override
-```
-
-**Best for:**
-
-- Varied request sizes
-- Fair usage billing
-- Cost optimization
-
-### 3. Model-Based Pricing
-
-Dynamic pricing based on model costs:
-
-```bash
-FIXED_PRICING=false
-EXCHANGE_FEE=1.005      # 0.5% exchange fee
-UPSTREAM_PROVIDER_FEE=1.05  # 5% provider fee
-```
-
-**Best for:**
-
-- Multiple models
-- Market-based pricing
-- Automatic updates
-
-## Model Configuration
-
-### Default Models
-
-Routstr includes pricing for popular models:
-
-| Model | Input ($/1K) | Output ($/1K) | Context | Notes |
-|-------|--------------|---------------|---------|-------|
-| gpt-3.5-turbo | $0.0015 | $0.002 | 16K | Fast, economical |
-| gpt-4 | $0.03 | $0.06 | 8K | Advanced reasoning |
-| gpt-4-turbo | $0.01 | $0.03 | 128K | Large context |
-| claude-3-opus | $0.015 | $0.075 | 200K | Best quality |
-| claude-3-sonnet | $0.003 | $0.015 | 200K | Balanced |
-| llama-2-70b | $0.0007 | $0.0009 | 4K | Open source |
-
-### Custom Models File
-
-Create `models.json` to override defaults:
-
-```json
-{
-  "models": [
-    {
-      "id": "gpt-4-vision",
-      "name": "GPT-4 Vision",
-      "pricing": {
-        "prompt": "0.00003",
-        "completion": "0.00006",
-        "request": "0",
-        "image": "0.00255"
-      },
-      "context_length": 128000,
-      "supports_vision": true
-    },
-    {
-      "id": "custom-model",
-      "name": "My Custom Model",
-      "pricing": {
-        "prompt": "0.001",
-        "completion": "0.002",
-        "request": "0.0001"
-      },
-      "context_length": 8192
-    }
-  ]
-}
-```
-
-### Auto-updating Models
-
-Fetch latest models from OpenRouter:
-
-```bash
-# Update models from API
-python scripts/models_meta.py
-
-# Or manually
-curl https://openrouter.ai/api/v1/models > models.json
-```
-
-## Cost Calculation
-
-### Understanding the Formula
-
-```
-Base Cost = (Input Tokens × Input Rate) + (Output Tokens × Output Rate) + Request Fee
-
-Bitcoin Price = Current BTC/USD rate (e.g., $50,000)
-Sats Cost = (Base Cost / Bitcoin Price) × 100,000,000
-
-Final Cost = Sats Cost × Exchange Fee × Provider Fee
-```
-
-### Example Calculations
-
-**Example 1: Simple Chat (gpt-3.5-turbo)**
-
-```
-Input: 50 tokens
-Output: 150 tokens
-Model rates: $0.0015/1K input, $0.002/1K output
-
-USD Cost = (50/1000 × 0.0015) + (150/1000 × 0.002)
-         = $0.000075 + $0.0003
-         = $0.000375
-
-At $50,000/BTC: 0.75 sats
-With 5.5% total fees: 0.79 sats
-```
-
-**Example 2: Large Context (gpt-4)**
-
-```
-Input: 2,000 tokens
-Output: 500 tokens
-Model rates: $0.03/1K input, $0.06/1K output
-
-USD Cost = (2000/1000 × 0.03) + (500/1000 × 0.06)
-         = $0.06 + $0.03
-         = $0.09
-
-At $50,000/BTC: 180 sats
-With 5.5% total fees: 190 sats
-```
-
-**Example 3: Image Generation (dall-e-3)**
-
-```
-Model: dall-e-3
-Size: 1024x1024
-Quality: standard
-Cost: $0.04 per image
-
-At $50,000/BTC: 80 sats
-With 5.5% fees: 84 sats
-```
-
-## Fee Structure
-
-### Exchange Fee
-
-Covers Bitcoin/USD conversion costs:
-
-```bash
-EXCHANGE_FEE=1.005  # 0.5% default
-```
-
-Factors:
-
-- Exchange rate volatility
-- Conversion costs
-- Price update frequency
-
-### Provider Fee
-
-Node operator's margin:
-
-```bash
-UPSTREAM_PROVIDER_FEE=1.05  # 5% default
-```
-
-Covers:
-
-- Infrastructure costs
-- Maintenance
-- Support
-- Profit margin
-
-### Calculating Total Fees
-
-```
-Total Multiplier = EXCHANGE_FEE × UPSTREAM_PROVIDER_FEE
-Example: 1.005 × 1.05 = 1.05525 (5.525% total)
-```
-
-## Special Pricing
-
-### Image Models
-
-Image generation uses per-image pricing:
-
-| Model | Size | Quality | Price |
-|-------|------|---------|-------|
-| dall-e-2 | 256x256 | - | $0.016 |
-| dall-e-2 | 512x512 | - | $0.018 |
-| dall-e-2 | 1024x1024 | - | $0.02 |
-| dall-e-3 | 1024x1024 | standard | $0.04 |
-| dall-e-3 | 1024x1024 | hd | $0.08 |
-| dall-e-3 | 1024x1792 | standard | $0.08 |
-| dall-e-3 | 1024x1792 | hd | $0.12 |
-
-### Audio Models
-
-Audio pricing by duration:
-
-| Model | Type | Price |
-|-------|------|-------|
-| whisper-1 | Transcription | $0.006/minute |
-| whisper-1 | Translation | $0.006/minute |
-| tts-1 | Text-to-speech | $0.015/1K chars |
-| tts-1-hd | HD speech | $0.03/1K chars |
-
-### Embedding Models
-
-Lower costs for embeddings:
-
-| Model | Price/1K tokens |
-|-------|-----------------|
-| text-embedding-3-small | $0.00002 |
-| text-embedding-3-large | $0.00013 |
-| text-embedding-ada-002 | $0.0001 |
-
-## Monitoring Costs
-
-### Per-Request Tracking
-
-Each API response includes usage data:
-
-```json
-{
-  "usage": {
-    "prompt_tokens": 50,
-    "completion_tokens": 150,
-    "total_tokens": 200
-  },
-  "x-routstr-cost": {
-    "sats": 79,
-    "usd": 0.000375,
-    "breakdown": {
-      "prompt_cost": 15,
-      "completion_cost": 60,
-      "fees": 4
-    }
-  }
-}
-```
-
-### Daily Summaries
-
-View in admin dashboard:
-
-- Total requests
-- Token usage by model
-- Cost distribution
-- Trending patterns
-
-### Cost Alerts
-
-Set up notifications:
-
-```python
-# Example monitoring script
-def check_daily_spend(api_key):
-    balance_start = get_balance(api_key, "00:00")
-    balance_now = get_balance(api_key)
-    spent = balance_start - balance_now
-    
-    if spent > DAILY_LIMIT:
-        send_alert(f"Daily spend exceeded: {spent} sats")
-```
-
-## Optimization Strategies
-
-### Model Selection
-
-Choose the right model for each task:
-
-| Task | Recommended Model | Why |
-|------|-------------------|-----|
-| Simple Q&A | gpt-3.5-turbo | Fast, cheap, sufficient |
-| Code generation | gpt-4 | Better reasoning |
-| Summarization | claude-3-haiku | Good balance |
-| Creative writing | claude-3-opus | Best quality |
-| Embeddings | text-embedding-3-small | Optimized for vectors |
-
-### Prompt Engineering
-
-Reduce costs with efficient prompts:
-
-```python
-# Expensive
-prompt = """
-You are an AI assistant. Your task is to help users.
-Please provide detailed, comprehensive answers.
-Now, answer this question: What is 2+2?
-"""
-
-# Economical
-prompt = "Calculate: 2+2"
-```
-
-### Caching Strategies
-
-Implement smart caching:
-
-```python
-# Cache embedding results
-@lru_cache(maxsize=1000)
-def get_embedding(text):
-    return client.embeddings.create(
-        model="text-embedding-3-small",
-        input=text
-    )
-
-# Cache common responses
-COMMON_RESPONSES = {
-    "greeting": "Hello! How can I help you?",
-    "goodbye": "Goodbye! Have a great day!"
-}
-```
-
-### Batch Processing
-
-Process multiple items efficiently:
-
-```python
-# Instead of multiple calls
-for item in items:
-    response = client.chat.completions.create(...)
-
-# Use single call with formatted prompt
-prompt = "\n".join([f"{i+1}. {item}" for i, item in enumerate(items)])
-response = client.chat.completions.create(
-    messages=[{"role": "user", "content": f"Process these items:\n{prompt}"}]
-)
-```
-
-## Custom Pricing Rules
-
-### Time-Based Pricing
-
-Implement off-peak discounts:
-
-```python
-def calculate_multiplier():
-    hour = datetime.now().hour
-    if 2 <= hour <= 6:  # 2 AM - 6 AM
-        return 0.8  # 20% discount
-    elif 18 <= hour <= 22:  # 6 PM - 10 PM
-        return 1.2  # 20% premium
-    return 1.0
-```
-
-### Model-Specific Rules
-
-Custom pricing logic:
-
-```python
-def adjust_model_price(model, base_price):
-    # Premium for latest models
-    if "turbo" in model or "latest" in model:
-        return base_price * 1.1
-    
-    # Discount for older models
-    if "legacy" in model:
-        return base_price * 0.8
-    
-    return base_price
-```
-
-## Pricing Transparency
-
-### Public Pricing Page
-
-Display current rates:
-
-```html
-<!-- Available at /pricing -->
-<table>
-  <tr>
-    <th>Model</th>
-    <th>Input (sats/1K)</th>
-    <th>Output (sats/1K)</th>
-  </tr>
-  <!-- Dynamically generated from models.json -->
-</table>
-```
-
-### Cost Estimation API
-
-Provide cost estimates:
-
-```bash
-POST /v1/estimate
-{
-  "model": "gpt-4",
-  "prompt_tokens": 500,
-  "max_tokens": 200
-}
-
-Response:
-{
-  "estimated_cost_sats": 45,
-  "breakdown": {
-    "prompt": 30,
-    "completion": 12,
-    "fees": 3
-  }
-}
-```
-
-## Troubleshooting
-
-### Pricing Mismatches
-
-**Issue**: Costs don't match expectations
-
-- Check current BTC/USD rate
-- Verify fee settings
-- Review model configuration
-
-**Issue**: Models not found
-
-- Update models.json
-- Check model ID spelling
-- Verify upstream support
-
-### Fee Calculations
-
-**Issue**: Fees seem too high
-
-- Review EXCHANGE_FEE setting
-- Check UPSTREAM_PROVIDER_FEE
-- Calculate total multiplier
-
-## Next Steps
-
-- [API Reference](../api/overview.md) - Technical details
-- [Custom Pricing](../advanced/custom-pricing.md) - Advanced configuration
-- [Contributing](../contributing/setup.md) - Help improve Routstr

docs/user-guide/payment-flow.md

@@ -1,373 +0,0 @@
-# Payment Flow
-
-Understanding how payments work in Routstr is key to using the system effectively. This guide explains the payment process in detail.
-
-## Overview
-
-Routstr uses a pre-funded account model where:
-
-1. Users deposit eCash tokens to create an API key
-2. Each API request deducts from the balance
-3. Users can withdraw remaining balance as eCash
-
-## Creating an API Key
-
-### Step 1: Obtain eCash Token
-
-Get a Cashu token from any compatible source:
-
-**Option A: From a Cashu Wallet**
-
-```bash
-# Example: Creating a 10,000 sat token
-cashu send 10000
-```
-
-**Option B: Lightning Invoice**
-
-```bash
-# Some mints support direct Lightning deposits
-curl -X POST https://mint.example.com/v1/mint/quote/bolt11 \
-  -d '{"amount": 10000, "unit": "sat"}'
-```
-
-**Option C: Test Tokens**
-
-```bash
-# Get test tokens from testnet mints
-# Check mint documentation for faucets
-```
-
-### Step 2: Create API Key
-
-**Note: The POST /v1/wallet/create endpoint is coming soon. Currently, you can use Cashu tokens directly as API keys in the Authorization header.**
-
-Send your token to Routstr:
-
-```bash
-curl -X POST https://api.routstr.com/v1/wallet/create \
-  -H "Content-Type: application/json" \
-  -d '{
-    "cashu_token": "cashuAeyJ0b2tlbiI6W3sibWludCI6Imh0dHBzOi8vbWlu..."
-  }'
-```
-
-**Request Parameters:**
-
-- `cashu_token` (required): The eCash token to deposit
-
-**Response:**
-
-```json
-{
-  "api_key": "sk-1234567890abcdef",
-  "balance": 10000000,
-  "created_at": "2024-01-01T00:00:00Z"
-}
-```
-
-### Step 3: Verify Balance
-
-Check your key's balance:
-
-```bash
-curl -X GET https://api.routstr.com/v1/wallet/balance \
-  -H "Authorization: Bearer sk-1234567890abcdef"
-```
-
-Response:
-
-```json
-{
-  "balance": 10000000,
-  "total_deposited": 10000000,
-  "total_spent": 0,
-  "last_used": null
-}
-```
-
-## Making API Requests
-
-### Cost Calculation
-
-Costs are calculated based on:
-
-1. **Request Type**
-   - Chat completions
-   - Embeddings
-   - Image generation
-   - Audio processing
-
-2. **Token Usage**
-   - Input tokens (prompt)
-   - Output tokens (response)
-   - Model-specific rates
-
-3. **Additional Costs**
-   - Base request fee
-   - Image generation fees
-   - Audio processing time
-
-### Example: Chat Completion
-
-```python
-import openai
-
-client = openai.OpenAI(
-    api_key="sk-1234567890abcdef",
-    base_url="https://api.routstr.com/v1"
-)
-
-# Make request
-response = client.chat.completions.create(
-    model="gpt-3.5-turbo",
-    messages=[
-        {"role": "user", "content": "Hello, how are you?"}
-    ]
-)
-
-# Check usage
-print(f"Input tokens: {response.usage.prompt_tokens}")
-print(f"Output tokens: {response.usage.completion_tokens}")
-print(f"Total tokens: {response.usage.total_tokens}")
-```
-
-### Cost Breakdown
-
-For the above request:
-
-```
-Model: gpt-3.5-turbo
-Input tokens: 13
-Output tokens: 27
-Model rates: $0.0015/1K input, $0.002/1K output
-
-USD Cost = (13/1000 * 0.0015) + (27/1000 * 0.002) = $0.0000735
-BTC/USD Rate: $50,000
-BTC Cost = 0.0000735 / 50000 = 0.00000000147 BTC = 147 sats
-With fees (5%): 154 sats
-
-Final cost: 154 sats
-```
-
-## Balance Management
-
-### Monitoring Usage
-
-Track your usage in real-time:
-
-```bash
-# Get current balance
-curl -X GET https://api.routstr.com/v1/wallet/balance \
-  -H "Authorization: Bearer your-api-key"
-
-# View recent transactions (through admin dashboard)
-# Access at https://api.routstr.com/admin/
-```
-
-### Low Balance Handling
-
-When balance is insufficient:
-
-```json
-{
-  "error": {
-    "type": "insufficient_balance",
-    "message": "Insufficient balance. Current: 100 sats, Required: 154 sats",
-    "code": "payment_required"
-  }
-}
-```
-
-### Topping Up
-
-Add funds to existing key:
-
-```bash
-curl -X POST https://api.routstr.com/v1/wallet/topup \
-  -H "Authorization: Bearer your-api-key" \
-  -H "Content-Type: application/json" \
-  -d '{
-    "cashu_token": "cashuAeyJ0b2..."
-  }'
-```
-
-## Withdrawing Balance
-
-### Via Admin Dashboard
-
-1. Navigate to `/admin/`
-2. Enter admin password
-3. Find your API key
-4. Click "Withdraw"
-5. Receive eCash token
-
-### Via API (if enabled)
-
-```bash
-curl -X POST https://api.routstr.com/v1/wallet/withdraw \
-  -H "Authorization: Bearer your-api-key" \
-  -H "Content-Type: application/json" \
-  -d '{
-    "amount": 5000,
-    "mint": "https://mint.minibits.cash/Bitcoin"
-  }'
-```
-
-## Payment Security
-
-### Token Validation
-
-Routstr validates tokens by:
-
-1. Checking signature validity
-2. Verifying with the issuing mint
-3. Ensuring no double-spending
-4. Confirming sufficient value
-
-### Failed Payments
-
-Common failure reasons:
-
-- Invalid token signature
-- Already spent token
-- Untrusted mint
-- Network issues with mint
-
-### Refund Policy
-
-- Unused balance can be withdrawn anytime
-- Expired keys with balance can be refunded
-- Node operators may have additional policies
-
-## Advanced Features
-
-### Multi-Mint Support
-
-Routstr accepts tokens from multiple mints:
-
-```bash
-CASHU_MINTS=https://mint1.com,https://mint2.com,https://mint3.com
-```
-
-Benefits:
-
-- Redundancy if one mint is down
-- User choice of mints
-- Geographic distribution
-
-### Automatic Payouts
-
-Configure automatic Lightning payouts:
-
-```bash
-RECEIVE_LN_ADDRESS=satoshi@getalby.com
-```
-
-When enabled:
-
-- Balances above threshold are swept
-- Converted to Lightning payments
-- Sent to configured address
-
-### Per-Request Payments (Coming Soon)
-
-Future support for Nut-24 headers:
-
-```bash
-curl -X POST https://api.routstr.com/v1/chat/completions \
-  -H "x-cashu: cashuAeyJ0..." \
-  -H "Content-Type: application/json" \
-  -d '{...}'
-```
-
-Response includes change:
-
-```
-HTTP/1.1 200 OK
-x-cashu: cashuAeyJjaGFuZ2Ui...
-```
-
-## Best Practices
-
-### API Key Management
-
-1. **Separate Keys per Application**
-   - Easier tracking
-   - Better security
-   - Independent budgets
-
-2. **Set Expiration Dates**
-   - Automatic cleanup
-   - Security improvement
-   - Budget control
-
-3. **Monitor Balances**
-   - Set up alerts
-   - Regular checks
-   - Usage analytics
-
-### Cost Optimization
-
-1. **Choose Appropriate Models**
-   - Smaller models for simple tasks
-   - Larger models only when needed
-
-2. **Optimize Prompts**
-   - Concise, clear instructions
-   - Avoid unnecessary tokens
-
-3. **Use Streaming**
-   - Early termination possible
-   - Better user experience
-
-### Security
-
-1. **Secure Storage**
-   - Environment variables
-   - Secrets management
-   - Never in code
-
-2. **Network Security**
-   - Always use HTTPS
-   - Verify certificates
-   - Consider Tor for privacy
-
-3. **Regular Rotation**
-   - Change keys periodically
-   - Withdraw unused funds
-   - Audit usage logs
-
-## Troubleshooting
-
-### Payment Rejected
-
-**Error:** "Invalid token"
-
-- Check token format
-- Verify mint is trusted
-- Ensure not already spent
-
-**Error:** "Insufficient value"
-
-- Token value too low
-- Check current pricing
-- Add larger token
-
-### Balance Discrepancies
-
-- Allow for price fluctuations
-- Check model pricing updates
-- Review transaction history
-
-### Mint Issues
-
-- Try different mint from list
-- Check mint status
-- Contact mint operator
-
-## Next Steps
-
-- [Using the API](using-api.md) - Integration guide
-- [Admin Dashboard](admin-dashboard.md) - Account management
-- [Models & Pricing](models-pricing.md) - Cost details

docs/user-guide/using-api.md

@@ -1,545 +0,0 @@
-# Using the API
-
-This guide shows how to integrate Routstr with your applications using various programming languages and tools.
-
-## API Compatibility
-
-Routstr maintains full compatibility with the OpenAI API, meaning:
-
-- Existing OpenAI client libraries work without modification
-- Only the base URL and API key need to change
-- All parameters and responses match OpenAI's format
-
-## Basic Setup
-
-### Python
-
-Using the official OpenAI Python library:
-
-```python
-from openai import OpenAI
-
-# Initialize client with Routstr endpoint
-client = OpenAI(
-    api_key="sk-...",
-    base_url="https://api.routstr.com/v1"
-)
-
-# Use exactly like OpenAI
-response = client.chat.completions.create(
-    model="gpt-3.5-turbo",
-    messages=[
-        {"role": "system", "content": "You are a helpful assistant."},
-        {"role": "user", "content": "Hello!"}
-    ]
-)
-
-print(response.choices[0].message.content)
-```
-
-### JavaScript/TypeScript
-
-Using the official OpenAI Node.js library:
-
-```javascript
-import OpenAI from 'openai';
-
-// Initialize client
-const openai = new OpenAI({
-    apiKey: 'sk-...',
-    baseURL: 'https://api.routstr.com/v1'
-});
-
-// Make a request
-async function main() {
-    const completion = await openai.chat.completions.create({
-        model: 'gpt-3.5-turbo',
-        messages: [
-            { role: 'system', content: 'You are a helpful assistant.' },
-            { role: 'user', content: 'Hello!' }
-        ]
-    });
-    
-    console.log(completion.choices[0].message.content);
-}
-
-main();
-```
-
-### cURL
-
-Direct HTTP requests:
-
-```bash
-curl https://api.routstr.com/v1/chat/completions \
-  -H "Content-Type: application/json" \
-  -H "Authorization: Bearer sk-..." \
-  -d '{
-    "model": "gpt-3.5-turbo",
-    "messages": [
-      {"role": "user", "content": "Hello!"}
-    ]
-  }'
-```
-
-## Common Use Cases
-
-### Chat Completions
-
-Standard chat with conversation history:
-
-```python
-messages = []
-
-def chat(user_input):
-    # Add user message
-    messages.append({"role": "user", "content": user_input})
-    
-    # Get AI response
-    response = client.chat.completions.create(
-        model="gpt-3.5-turbo",
-        messages=messages,
-        temperature=0.7,
-        max_tokens=150
-    )
-    
-    # Add AI response to history
-    ai_message = response.choices[0].message
-    messages.append({"role": "assistant", "content": ai_message.content})
-    
-    return ai_message.content
-
-# Usage
-print(chat("What's the weather like?"))
-print(chat("How should I dress?"))  # Maintains context
-```
-
-### Streaming Responses
-
-For real-time output:
-
-```python
-stream = client.chat.completions.create(
-    model="gpt-3.5-turbo",
-    messages=[{"role": "user", "content": "Write a short story"}],
-    stream=True
-)
-
-for chunk in stream:
-    if chunk.choices[0].delta.content is not None:
-        print(chunk.choices[0].delta.content, end="", flush=True)
-```
-
-### Function Calling
-
-Using OpenAI's function calling feature:
-
-```python
-tools = [{
-    "type": "function",
-    "function": {
-        "name": "get_weather",
-        "description": "Get current weather",
-        "parameters": {
-            "type": "object",
-            "properties": {
-                "location": {"type": "string"},
-                "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}
-            },
-            "required": ["location"]
-        }
-    }
-}]
-
-response = client.chat.completions.create(
-    model="gpt-4",
-    messages=[{"role": "user", "content": "What's the weather in Tokyo?"}],
-    tools=tools,
-    tool_choice="auto"
-)
-
-# Check if function was called
-if response.choices[0].message.tool_calls:
-    tool_call = response.choices[0].message.tool_calls[0]
-    print(f"Function: {tool_call.function.name}")
-    print(f"Arguments: {tool_call.function.arguments}")
-```
-
-### Embeddings
-
-Generate text embeddings:
-
-```python
-response = client.embeddings.create(
-    model="text-embedding-3-small",
-    input="The quick brown fox jumps over the lazy dog"
-)
-
-embedding = response.data[0].embedding
-print(f"Embedding dimension: {len(embedding)}")
-```
-
-### Image Generation
-
-Create images with DALL-E:
-
-```python
-response = client.images.generate(
-    model="dall-e-3",
-    prompt="A futuristic city with flying cars",
-    size="1024x1024",
-    quality="standard",
-    n=1
-)
-
-image_url = response.data[0].url
-print(f"Image URL: {image_url}")
-```
-
-### Audio Transcription
-
-Convert speech to text:
-
-```python
-with open("audio.mp3", "rb") as audio_file:
-    response = client.audio.transcriptions.create(
-        model="whisper-1",
-        file=audio_file,
-        response_format="text"
-    )
-    
-print(response.text)
-```
-
-## Error Handling
-
-### Balance Errors
-
-Handle insufficient balance gracefully:
-
-```python
-try:
-    response = client.chat.completions.create(
-        model="gpt-4",
-        messages=[{"role": "user", "content": "Hello"}]
-    )
-except Exception as e:
-    if "insufficient_balance" in str(e):
-        print("Low balance! Please top up your API key.")
-        # Implement top-up logic
-    else:
-        raise
-```
-
-### Rate Limiting
-
-Implement exponential backoff:
-
-```python
-import time
-from typing import Optional
-
-def make_request_with_retry(
-    func, 
-    max_retries: int = 3,
-    initial_delay: float = 1.0
-) -> Optional[any]:
-    for attempt in range(max_retries):
-        try:
-            return func()
-        except Exception as e:
-            if "rate_limit" in str(e) and attempt < max_retries - 1:
-                delay = initial_delay * (2 ** attempt)
-                print(f"Rate limited. Waiting {delay}s...")
-                time.sleep(delay)
-            else:
-                raise
-    return None
-```
-
-### Connection Errors
-
-Handle network issues:
-
-```python
-import httpx
-
-# Configure timeout and retries
-client = OpenAI(
-    api_key="sk-...",
-    base_url="https://your-node.com/v1",
-    timeout=httpx.Timeout(60.0, connect=5.0),
-    max_retries=2
-)
-```
-
-## Advanced Features
-
-### Using Tor
-
-Route requests through Tor for privacy:
-
-```python
-import httpx
-
-# Configure Tor proxy
-proxies = {
-    "http://": "socks5://127.0.0.1:9050",
-    "https://": "socks5://127.0.0.1:9050"
-}
-
-http_client = httpx.Client(proxies=proxies)
-
-client = OpenAI(
-    api_key="sk-...",
-    base_url="http://your-onion-address.onion/v1",
-    http_client=http_client
-)
-```
-
-### Custom Headers
-
-Add custom headers if needed:
-
-```python
-import httpx
-
-class CustomClient(httpx.Client):
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.headers["X-Custom-Header"] = "value"
-
-client = OpenAI(
-    api_key="sk-...",
-    base_url="https://your-node.com/v1",
-    http_client=CustomClient()
-)
-```
-
-### Azure OpenAI compatibility
-
-To use Azure OpenAI through Routstr with minimal changes:
-
-- Set `UPSTREAM_BASE_URL` to your Azure deployments URL, for example: `https://<resource>.openai.azure.com/openai/deployments/<deployment>`
-- Set `CHAT_COMPLETIONS_API_VERSION=2024-05-01-preview`
-
-When this env var is set, Routstr automatically appends `api-version=2024-05-01-preview` to all upstream `/chat/completions` requests.
-
-### Async Operations
-
-For high-performance applications:
-
-```python
-import asyncio
-from openai import AsyncOpenAI
-
-async_client = AsyncOpenAI(
-    api_key="sk-...",
-    base_url="https://your-node.com/v1"
-)
-
-async def process_messages(messages):
-    tasks = []
-    for msg in messages:
-        task = async_client.chat.completions.create(
-            model="gpt-3.5-turbo",
-            messages=[{"role": "user", "content": msg}]
-        )
-        tasks.append(task)
-    
-    responses = await asyncio.gather(*tasks)
-    return [r.choices[0].message.content for r in responses]
-
-# Run async
-messages = ["Hello", "How are you?", "What's 2+2?"]
-results = asyncio.run(process_messages(messages))
-```
-
-## Best Practices
-
-### 1. Environment Variables
-
-Never hardcode API keys:
-
-```python
-import os
-from openai import OpenAI
-
-client = OpenAI(
-    api_key=os.getenv("ROUTSTR_API_KEY"),
-    base_url=os.getenv("ROUTSTR_BASE_URL", "https://api.routstr.com/v1")
-)
-```
-
-### 2. Error Logging
-
-Implement comprehensive logging:
-
-```python
-import logging
-
-logging.basicConfig(level=logging.INFO)
-logger = logging.getLogger(__name__)
-
-try:
-    response = client.chat.completions.create(...)
-    logger.info(f"Request successful. Tokens used: {response.usage.total_tokens}")
-except Exception as e:
-    logger.error(f"API request failed: {e}")
-    raise
-```
-
-### 3. Cost Tracking
-
-Monitor your usage:
-
-```python
-class UsageTracker:
-    def __init__(self):
-        self.total_tokens = 0
-        self.total_requests = 0
-    
-    def track(self, response):
-        self.total_tokens += response.usage.total_tokens
-        self.total_requests += 1
-        
-        # Estimate cost (example rates)
-        cost_per_1k = 0.002  # $0.002 per 1K tokens
-        estimated_cost = (self.total_tokens / 1000) * cost_per_1k
-        
-        logger.info(f"Total usage: {self.total_tokens} tokens, "
-                   f"${estimated_cost:.4f} (~{estimated_cost * 50000:.0f} sats)")
-
-tracker = UsageTracker()
-response = client.chat.completions.create(...)
-tracker.track(response)
-```
-
-### 4. Caching Responses
-
-Reduce costs with intelligent caching:
-
-```python
-import hashlib
-import json
-from functools import lru_cache
-
-@lru_cache(maxsize=100)
-def cached_completion(prompt: str, model: str = "gpt-3.5-turbo"):
-    response = client.chat.completions.create(
-        model=model,
-        messages=[{"role": "user", "content": prompt}],
-        temperature=0  # Deterministic for caching
-    )
-    return response.choices[0].message.content
-
-# Repeated calls with same prompt use cache
-result1 = cached_completion("What is 2+2?")
-result2 = cached_completion("What is 2+2?")  # From cache, no API call
-```
-
-## Testing
-
-### Mock Responses
-
-For development without spending sats:
-
-```python
-class MockOpenAI:
-    class Completions:
-        def create(self, **kwargs):
-            return type('Response', (), {
-                'choices': [type('Choice', (), {
-                    'message': type('Message', (), {
-                        'content': 'Mock response'
-                    })()
-                })],
-                'usage': type('Usage', (), {
-                    'total_tokens': 10
-                })()
-            })()
-    
-    def __init__(self):
-        self.chat = type('Chat', (), {
-            'completions': self.Completions()
-        })()
-
-# Use mock in tests
-if os.getenv('TESTING'):
-    client = MockOpenAI()
-else:
-    client = OpenAI(...)
-```
-
-### Integration Tests
-
-Test your Routstr integration:
-
-```python
-def test_routstr_connection():
-    try:
-        # Test models endpoint
-        models = client.models.list()
-        assert len(models.data) > 0
-        
-        # Test simple completion
-        response = client.chat.completions.create(
-            model="gpt-3.5-turbo",
-            messages=[{"role": "user", "content": "test"}],
-            max_tokens=5
-        )
-        assert response.choices[0].message.content
-        
-        print("✅ Routstr integration working!")
-        return True
-    except Exception as e:
-        print(f"❌ Integration test failed: {e}")
-        return False
-```
-
-## Troubleshooting
-
-### Common Issues
-
-**SSL Certificate Errors**
-
-```python
-# For development only - not for production!
-import ssl
-import httpx
-
-client = OpenAI(
-    api_key="sk-...",
-    base_url="https://localhost:8000/v1",
-    http_client=httpx.Client(verify=False)
-)
-```
-
-**Timeout Issues**
-
-```python
-# Increase timeout for slow connections
-client = OpenAI(
-    api_key="sk-...",
-    base_url="https://your-node.com/v1",
-    timeout=httpx.Timeout(120.0)  # 2 minutes
-)
-```
-
-**Debugging Requests**
-
-```python
-import logging
-import httpx
-
-# Enable debug logging
-logging.basicConfig(level=logging.DEBUG)
-httpx_logger = logging.getLogger("httpx")
-httpx_logger.setLevel(logging.DEBUG)
-```
-
-## Next Steps
-
-- [Admin Dashboard](admin-dashboard.md) - Manage your account
-- [Models & Pricing](models-pricing.md) - Understanding costs
-- [API Reference](../api/overview.md) - Technical details

examples/create_child_keys.py

@@ -0,0 +1,45 @@
+import json
+import sys
+
+import httpx
+
+
+def create_child_keys(base_url: str, api_key: str, count: int = 3) -> list[str]:
+    headers = {"Authorization": f"Bearer {api_key}"}
+
+    print(f"Requesting {count} child keys from {base_url}...")
+
+    child_keys = []
+
+    for i in range(count):
+        try:
+            response = httpx.post(f"{base_url}/v1/balance/child-key", headers=headers)
+            if response.status_code == 200:
+                data = response.json()
+                child_keys.append(data["api_key"])
+                print(
+                    f"  [{i + 1}] Created: {data['api_key']} (Cost: {data['cost_msats']} msats)"
+                )
+            else:
+                print(f"  [{i + 1}] Failed: {response.status_code} - {response.text}")
+        except Exception as e:
+            print(f"  [{i + 1}] Error: {str(e)}")
+
+    return child_keys
+
+
+if __name__ == "__main__":
+    if len(sys.argv) < 2:
+        print("Usage: python create_child_keys.py <api_key_or_cashu_token> [base_url]")
+        sys.exit(1)
+
+    auth_key = sys.argv[1]
+    base_url = sys.argv[2] if len(sys.argv) > 2 else "http://localhost:8000"
+
+    keys = create_child_keys(base_url, auth_key)
+
+    if keys:
+        print("\nSuccessfully created child keys:")
+        print(json.dumps(keys, indent=2))
+    else:
+        print("\nNo child keys were created.")

migrations/versions/06f81c0fc88d_add_key_management_and_reset_fields_to_.py

@@ -0,0 +1,37 @@
+"""add key management and reset fields to api_keys
+
+Revision ID: 06f81c0fc88d
+Revises: c2d3e4f5a6b7
+Create Date: 2026-02-04 22:44:03.311983
+"""
+
+import sqlalchemy as sa
+import sqlmodel
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "06f81c0fc88d"
+down_revision = "c2d3e4f5a6b7"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column("api_keys", sa.Column("balance_limit", sa.Integer(), nullable=True))
+    op.add_column(
+        "api_keys",
+        sa.Column(
+            "balance_limit_reset", sqlmodel.sql.sqltypes.AutoString(), nullable=True
+        ),
+    )
+    op.add_column(
+        "api_keys", sa.Column("balance_limit_reset_date", sa.Integer(), nullable=True)
+    )
+    op.add_column("api_keys", sa.Column("validity_date", sa.Integer(), nullable=True))
+
+
+def downgrade() -> None:
+    op.drop_column("api_keys", "validity_date")
+    op.drop_column("api_keys", "balance_limit_reset_date")
+    op.drop_column("api_keys", "balance_limit_reset")
+    op.drop_column("api_keys", "balance_limit")

migrations/versions/a86e5348850b_.py

@@ -0,0 +1,42 @@
+"""
+
+Revision ID: a86e5348850b
+Revises: b9667ffc5701
+Create Date: 2026-01-10 18:57:48.475781
+"""
+
+import sqlalchemy as sa
+import sqlmodel
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "a86e5348850b"
+down_revision = "b9667ffc5701"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # Use batch_alter_table for SQLite compatibility
+    with op.batch_alter_table("api_keys", schema=None) as batch_op:
+        batch_op.add_column(
+            sa.Column(
+                "parent_key_hash", sqlmodel.sql.sqltypes.AutoString(), nullable=True
+            )
+        )
+        batch_op.create_index(
+            batch_op.f("ix_api_keys_parent_key_hash"), ["parent_key_hash"], unique=False
+        )
+        batch_op.create_foreign_key(
+            "fk_api_keys_parent_key_hash",
+            "api_keys",
+            ["parent_key_hash"],
+            ["hashed_key"],
+        )
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("api_keys", schema=None) as batch_op:
+        batch_op.drop_constraint("fk_api_keys_parent_key_hash", type_="foreignkey")
+        batch_op.drop_index(batch_op.f("ix_api_keys_parent_key_hash"))
+        batch_op.drop_column("parent_key_hash")

migrations/versions/c2d3e4f5a6b7_upstream_provider_base_url_api_key_unique.py

@@ -0,0 +1,118 @@
+"""make upstream provider base_url + api_key unique
+
+Revision ID: c2d3e4f5a6b7
+Revises: a86e5348850b
+Create Date: 2026-01-25 00:00:00.000000
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "c2d3e4f5a6b7"
+down_revision = "a86e5348850b"
+branch_labels = None
+depends_on = None
+
+
+def _recreate_table_sqlite(add_base_url_unique: bool) -> None:
+    conn = op.get_bind()
+    existing_tables = {
+        row[0]
+        for row in conn.exec_driver_sql(
+            "SELECT name FROM sqlite_master WHERE type='table'"
+        ).fetchall()
+    }
+    if "upstream_providers_old" in existing_tables:
+        if "upstream_providers" in existing_tables:
+            op.drop_table("upstream_providers_old")
+        else:
+            op.execute(
+                "ALTER TABLE upstream_providers_old RENAME TO upstream_providers"
+            )
+            existing_tables.add("upstream_providers")
+    if "upstream_providers" not in existing_tables:
+        return
+
+    constraints = [
+        sa.UniqueConstraint(
+            "base_url",
+            "api_key",
+            name="uq_upstream_providers_base_url_api_key",
+        )
+    ]
+    if add_base_url_unique:
+        constraints.append(
+            sa.UniqueConstraint("base_url", name="uq_upstream_providers_base_url")
+        )
+
+    op.execute("ALTER TABLE upstream_providers RENAME TO upstream_providers_old")
+    op.create_table(
+        "upstream_providers",
+        sa.Column(
+            "id", sa.Integer(), primary_key=True, nullable=False, autoincrement=True
+        ),
+        sa.Column("provider_type", sa.String(), nullable=False),
+        sa.Column("base_url", sa.String(), nullable=False),
+        sa.Column("api_key", sa.String(), nullable=False),
+        sa.Column("api_version", sa.String(), nullable=True),
+        sa.Column("enabled", sa.Boolean(), nullable=False),
+        sa.Column("provider_fee", sa.Float(), nullable=False, server_default="1.01"),
+        *constraints,
+    )
+    op.execute(
+        "INSERT INTO upstream_providers (id, provider_type, base_url, api_key, api_version, enabled, provider_fee) "
+        "SELECT id, provider_type, base_url, api_key, api_version, enabled, provider_fee "
+        "FROM upstream_providers_old"
+    )
+    op.drop_table("upstream_providers_old")
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    if conn.dialect.name == "sqlite":
+        _recreate_table_sqlite(add_base_url_unique=False)
+        return
+
+    inspector = sa.inspect(conn)
+    for constraint in inspector.get_unique_constraints("upstream_providers"):
+        name = constraint.get("name")
+        if constraint.get("column_names") == ["base_url"] and name:
+            op.drop_constraint(
+                name,
+                "upstream_providers",
+                type_="unique",
+            )
+    index_names = {idx["name"] for idx in inspector.get_indexes("upstream_providers")}
+    if "ix_upstream_providers_base_url" in index_names:
+        op.drop_index("ix_upstream_providers_base_url", table_name="upstream_providers")
+    op.create_unique_constraint(
+        "uq_upstream_providers_base_url_api_key",
+        "upstream_providers",
+        ["base_url", "api_key"],
+    )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    if conn.dialect.name == "sqlite":
+        _recreate_table_sqlite(add_base_url_unique=True)
+        return
+
+    op.drop_constraint(
+        "uq_upstream_providers_base_url_api_key",
+        "upstream_providers",
+        type_="unique",
+    )
+    op.create_unique_constraint(
+        "uq_upstream_providers_base_url",
+        "upstream_providers",
+        ["base_url"],
+    )
+    op.create_index(
+        "ix_upstream_providers_base_url",
+        "upstream_providers",
+        ["base_url"],
+        unique=True,
+    )

mkdocs.yml

@@ -77,29 +77,27 @@ extra:
 
 nav:
   - Home: index.md
-  - Getting Started:
-      - Overview: getting-started/overview.md
-      - Quick Start: getting-started/quickstart.md
-      - Docker Setup: getting-started/docker.md
-      - Configuration: getting-started/configuration.md
-  - User Guide:
-      - Introduction: user-guide/introduction.md
-      - Payment Flow: user-guide/payment-flow.md
-      - Using the API: user-guide/using-api.md
-      - Admin Dashboard: user-guide/admin-dashboard.md
-      - Models & Pricing: user-guide/models-pricing.md
-  - Contributing:
-      - Setup Development: contributing/setup.md
-      - Architecture: contributing/architecture.md
-      - Code Structure: contributing/code-structure.md
-      - Testing: contributing/testing.md
+  - Overview: overview.md
+  - Client Guide:
+      - Introduction: client/introduction.md
+      - Payment Flow: client/payments.md
+      - Integration: client/integration.md
+  - Provider Guide:
+      - Quick Start: provider/quickstart.md
+      - Dashboard: provider/dashboard.md
+      - Deployment: provider/deployment.md
+      - Configuration: provider/configuration.md
+      - Pricing: provider/pricing.md
+      - Advanced Pricing: provider/advanced-pricing.md
+      - Discovery: provider/discovery.md
+      - Tor Support: provider/tor.md
   - API Reference:
       - Overview: api/overview.md
       - Authentication: api/authentication.md
       - Endpoints: api/endpoints.md
       - Errors: api/errors.md
-  - Advanced:
-      - Tor Support: advanced/tor.md
-      - Nostr Discovery: advanced/nostr.md
-      - Custom Pricing: advanced/custom-pricing.md
-      - Migrations: advanced/migrations.md
+  - Contributing:
+      - Setup Development: contributing/setup.md
+      - Architecture: contributing/architecture.md
+      - Code Structure: contributing/code-structure.md
+      - Testing: contributing/testing.md

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "routstr"
-version = "0.2.1"
+version = "0.3.0"
 description = "Payment proxy for your LLM endpoint using cashu and nostr."
 readme = "README.md"
 requires-python = ">=3.11"

routstr/algorithm.py

@@ -84,93 +84,26 @@ def get_provider_penalty(provider: "BaseUpstreamProvider") -> float:
     return penalty
 
 
-def should_prefer_model(
-    candidate_model: "Model",
-    candidate_provider: "BaseUpstreamProvider",
-    current_model: "Model",
-    current_provider: "BaseUpstreamProvider",
-    alias: str,
-) -> bool:
-    """Determine if candidate model should replace current model for an alias.
-
-    This is the core decision function for model prioritization. It considers:
-    1. Alias matching quality (exact match vs. canonical slug match)
-    2. Model cost (lower is better)
-    3. Provider penalties (e.g., slight preference against OpenRouter)
-
-    Args:
-        candidate_model: The new model being considered
-        candidate_provider: Provider offering the candidate model
-        current_model: The currently selected model for this alias
-        current_provider: Provider offering the current model
-        alias: The model alias being mapped
-
-    Returns:
-        True if candidate should replace current, False otherwise
-    """
-
-    def get_base_model_id(model_id: str) -> str:
-        """Get base model ID by removing provider prefix."""
-        return model_id.split("/", 1)[1] if "/" in model_id else model_id
-
-    def alias_priority(model: "Model") -> int:
-        """Rank how strong the mapping of alias->model is.
-
-        Highest priority when alias exactly equals the model ID without provider prefix.
-        Next when alias equals canonical slug without prefix. Otherwise lowest.
-        """
-        model_base = get_base_model_id(model.id)
-        if model_base == alias:
-            return 3
-        if model.canonical_slug:
-            canonical_base = get_base_model_id(model.canonical_slug)
-            if canonical_base == alias:
-                return 2
-        return 1
-
-    candidate_alias_priority = alias_priority(candidate_model)
-    current_alias_priority = alias_priority(current_model)
-
-    # If candidate has better alias match, prefer it regardless of cost
-    if candidate_alias_priority > current_alias_priority:
-        return True
-
-    # If current has better alias match, keep it regardless of cost
-    if current_alias_priority > candidate_alias_priority:
-        return False
-
-    # Same alias priority - compare costs
-    candidate_cost = calculate_model_cost_score(candidate_model)
-    current_cost = calculate_model_cost_score(current_model)
-
-    # Apply provider penalties
-    candidate_adjusted = candidate_cost * get_provider_penalty(candidate_provider)
-    current_adjusted = current_cost * get_provider_penalty(current_provider)
-
-    # Prefer lower adjusted cost
-    should_replace = candidate_adjusted < current_adjusted
-
-    return should_replace
-
-
 def create_model_mappings(
     upstreams: list["BaseUpstreamProvider"],
     overrides_by_id: dict[str, tuple],
     disabled_model_ids: set[str],
-) -> tuple[dict[str, "Model"], dict[str, "BaseUpstreamProvider"], dict[str, "Model"]]:
+) -> tuple[
+    dict[str, "Model"], dict[str, list["BaseUpstreamProvider"]], dict[str, "Model"]
+]:
     """Create optimal model mappings based on cost and provider preferences.
 
     This is the main entry point for the algorithm. It processes all upstream providers
     and creates three mappings based on cost optimization:
 
     1. model_instances: alias -> Model (all model aliases mapped to their Model objects)
-    2. provider_map: alias -> UpstreamProvider (which provider to use for each alias)
+    2. provider_map: alias -> List[UpstreamProvider] (sorted list of providers for each alias)
     3. unique_models: base_id -> Model (unique models without provider prefixes)
 
     The algorithm:
     - Processes non-OpenRouter providers first (they're typically cheaper)
     - Then processes OpenRouter models (they can still win if cheaper)
-    - For each model alias, uses should_prefer_model() to select the best provider
+    - For each model alias, collects all candidates and sorts them by priority and cost.
 
     Args:
         upstreams: List of all upstream provider instances
@@ -183,8 +116,7 @@ def create_model_mappings(
     from .payment.models import _row_to_model
     from .upstream.helpers import resolve_model_alias
 
-    model_instances: dict[str, "Model"] = {}
-    provider_map: dict[str, "BaseUpstreamProvider"] = {}
+    candidates: dict[str, list[tuple["Model", "BaseUpstreamProvider"]]] = {}
     unique_models: dict[str, "Model"] = {}
 
     # Separate OpenRouter from other providers
@@ -202,24 +134,14 @@ def create_model_mappings(
         """Get base model ID by removing provider prefix."""
         return model_id.split("/", 1)[1] if "/" in model_id else model_id
 
-    def _maybe_set_alias(
+    def _add_candidate(
         alias: str, model: "Model", provider: "BaseUpstreamProvider"
     ) -> None:
-        """Set alias to model/provider if not set or if new model is preferred."""
+        """Add candidate model/provider for an alias."""
         alias_lower = alias.lower()
-        existing_model = model_instances.get(alias_lower)
-        if not existing_model:
-            # No existing mapping, set it
-            model_instances[alias_lower] = model
-            provider_map[alias_lower] = provider
-        else:
-            # Check if candidate should replace existing
-            existing_provider = provider_map[alias_lower]
-            if should_prefer_model(
-                model, provider, existing_model, existing_provider, alias
-            ):
-                model_instances[alias_lower] = model
-                provider_map[alias_lower] = provider
+        if alias_lower not in candidates:
+            candidates[alias_lower] = []
+        candidates[alias_lower].append((model, provider))
 
     def process_provider_models(
         upstream: "BaseUpstreamProvider", is_openrouter: bool = False
@@ -266,21 +188,55 @@ def create_model_mappings(
 
             # Try to set each alias
             for alias in aliases:
-                _maybe_set_alias(alias, model_to_use, upstream)
+                _add_candidate(alias, model_to_use, upstream)
 
-    # Process non-OpenRouter providers first (they're typically cheaper)
+    # Process non-OpenRouter providers first
     for upstream in other_upstreams:
         process_provider_models(upstream, is_openrouter=False)
 
-    # Process OpenRouter last - models only win if they're cheaper or better matched
+    # Process OpenRouter last
     if openrouter:
         process_provider_models(openrouter, is_openrouter=True)
 
-    # Log provider distribution
+    # Sort candidates and build final maps
+    model_instances: dict[str, "Model"] = {}
+    provider_map: dict[str, list["BaseUpstreamProvider"]] = {}
+
+    def alias_priority(model: "Model", alias: str) -> int:
+        """Rank how strong the mapping of alias->model is."""
+        model_base = get_base_model_id(model.id)
+        if model_base == alias:
+            return 3
+        if model.canonical_slug:
+            canonical_base = get_base_model_id(model.canonical_slug)
+            if canonical_base == alias:
+                return 2
+        return 1
+
+    for alias, items in candidates.items():
+        # Sort key: (priority DESC, cost ASC)
+        # Using negative cost for DESC sort overall to keep high priority first
+        def sort_key(item: tuple["Model", "BaseUpstreamProvider"]) -> tuple[int, float]:
+            model, provider = item
+            priority = alias_priority(model, alias)
+            cost = calculate_model_cost_score(model)
+            penalty = get_provider_penalty(provider)
+            adjusted_cost = cost * penalty
+            return (priority, -adjusted_cost)
+
+        items.sort(key=sort_key, reverse=True)
+
+        best_model, best_provider = items[0]
+        model_instances[alias] = best_model
+        provider_map[alias] = [p for _, p in items]
+
+    # Log provider distribution (using top provider for stats)
     provider_counts: dict[str, int] = {}
-    for provider in provider_map.values():
-        provider_name = getattr(provider, "upstream_name", "unknown")
-        provider_counts[provider_name] = provider_counts.get(provider_name, 0) + 1
+    for providers in provider_map.values():
+        if providers:
+            provider = providers[0]
+            provider_name = getattr(provider, "upstream_name", "unknown")
+            provider_counts[provider_name] = provider_counts.get(provider_name, 0) + 1
 
     logger.debug(
         f"Updated model mappings with ({len(unique_models)} unique models and {len(model_instances)} aliases)",

routstr/auth.py

@@ -1,10 +1,14 @@
+import asyncio
 import hashlib
 import math
+import random
+import time
+from datetime import datetime
 from typing import Optional
 
 from fastapi import HTTPException
 from sqlalchemy.exc import IntegrityError
-from sqlmodel import col, update
+from sqlmodel import col, select, update
 
 from .core import get_logger
 from .core.db import ApiKey, AsyncSession
@@ -24,16 +28,60 @@ logger = get_logger(__name__)
 # PREPAID_BALANCE = int(os.environ.get("PREPAID_BALANCE", "0")) * 1000  # Convert to msats
 
 
+async def check_and_reset_limit(key: ApiKey, session: AsyncSession) -> bool:
+    """Checks if a key's balance limit should be reset based on its policy."""
+    if key.balance_limit is not None and key.balance_limit_reset:
+        now = int(time.time())
+        reset_date = key.balance_limit_reset_date or 0
+        should_reset = False
+
+        if key.balance_limit_reset == "daily":
+            if (
+                datetime.fromtimestamp(now).date()
+                > datetime.fromtimestamp(reset_date).date()
+            ):
+                should_reset = True
+        elif key.balance_limit_reset == "weekly":
+            if (
+                datetime.fromtimestamp(now).isocalendar()[:2]
+                > datetime.fromtimestamp(reset_date).isocalendar()[:2]
+            ):
+                should_reset = True
+        elif key.balance_limit_reset == "monthly":
+            dt_now = datetime.fromtimestamp(now)
+            dt_reset = datetime.fromtimestamp(reset_date)
+            if dt_now.year > dt_reset.year or dt_now.month > dt_reset.month:
+                should_reset = True
+
+        if should_reset:
+            logger.info(
+                "Resetting balance limit for key",
+                extra={
+                    "key_hash": key.hashed_key[:8] + "...",
+                    "policy": key.balance_limit_reset,
+                    "old_spent": key.total_spent,
+                },
+            )
+            key.total_spent = 0
+            key.balance_limit_reset_date = now
+            session.add(key)
+            await session.flush()
+            return True
+    return False
+
+
 async def validate_bearer_key(
     bearer_key: str,
     session: AsyncSession,
     refund_address: Optional[str] = None,
     key_expiry_time: Optional[int] = None,
+    min_cost: int = 0,
 ) -> ApiKey:
     """
     Validates the provided API key using SQLModel.
     If it's a cashu key, it redeems it and stores its hash and balance.
     Otherwise checks if the hash of the key exists.
+    Includes a balance check against min_cost for limited keys.
     """
     logger.debug(
         "Starting bearer key validation",
@@ -43,6 +91,7 @@ async def validate_bearer_key(
             else bearer_key,
             "has_refund_address": bool(refund_address),
             "has_expiry_time": bool(key_expiry_time),
+            "min_cost": min_cost,
         },
     )
 
@@ -97,6 +146,50 @@ async def validate_bearer_key(
                     },
                 )
 
+            # Check and reset limit if needed
+            await check_and_reset_limit(existing_key, session)
+
+            # Early check: Billing balance check (Parent balance)
+            billing_key = await get_billing_key(existing_key, session)
+            if min_cost > 0 and billing_key.total_balance < min_cost:
+                logger.warning(
+                    "Insufficient billing balance during validation",
+                    extra={
+                        "key_hash": existing_key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                        "balance": billing_key.total_balance,
+                        "required": min_cost,
+                    },
+                )
+                raise HTTPException(
+                    status_code=402,
+                    detail={
+                        "error": {
+                            "message": f"Insufficient balance: {min_cost} mSats required for this model. {billing_key.total_balance} available.",
+                            "type": "insufficient_quota",
+                            "code": "insufficient_balance",
+                        }
+                    },
+                )
+
+            # Early check: Spending limit check (Child key limit)
+            if (
+                min_cost > 0
+                and existing_key.balance_limit is not None
+                and existing_key.total_spent + existing_key.reserved_balance + min_cost
+                > existing_key.balance_limit
+            ):
+                raise HTTPException(
+                    status_code=402,
+                    detail={
+                        "error": {
+                            "message": f"Balance limit exceeded: {existing_key.balance_limit} mSats limit. {existing_key.total_spent} already spent ({existing_key.reserved_balance} reserved), {min_cost} minimum required for this model.",
+                            "type": "insufficient_quota",
+                            "code": "balance_limit_exceeded",
+                        }
+                    },
+                )
+
             return existing_key
         else:
             logger.warning(
@@ -152,6 +245,19 @@ async def validate_bearer_key(
                         },
                     )
 
+                # Early check: Billing balance check
+                if min_cost > 0 and existing_key.total_balance < min_cost:
+                    raise HTTPException(
+                        status_code=402,
+                        detail={
+                            "error": {
+                                "message": f"Insufficient balance: {min_cost} mSats required for this model. {existing_key.total_balance} available.",
+                                "type": "insufficient_quota",
+                                "code": "insufficient_balance",
+                            }
+                        },
+                    )
+
                 return existing_key
 
             logger.info(
@@ -286,30 +392,57 @@ async def validate_bearer_key(
     )
 
 
+async def get_billing_key(key: ApiKey, session: AsyncSession) -> ApiKey:
+    """Returns the key that should be charged for the request."""
+    if key.parent_key_hash:
+        parent = await session.get(ApiKey, key.parent_key_hash)
+        if parent:
+            # We want to keep the total_requests and total_spent on the child key
+            # but use the balance and reserved_balance of the parent.
+            # However, pay_for_request updates reserved_balance and total_requests.
+            # To stay simple, we charge the parent's balance and update parent's total_requests.
+            return parent
+        else:
+            logger.error(
+                "Parent key not found for child key",
+                extra={
+                    "child_key_hash": key.hashed_key[:8] + "...",
+                    "parent_key_hash": key.parent_key_hash[:8] + "...",
+                },
+            )
+    return key
+
+
 async def pay_for_request(
     key: ApiKey, cost_per_request: int, session: AsyncSession
 ) -> int:
     """Process payment for a request."""
+    # Ensure cost_per_request is at least the minimum allowed request cost
+    cost_per_request = max(cost_per_request, settings.min_request_msat)
+
+    billing_key = await get_billing_key(key, session)
 
     logger.info(
         "Processing payment for request",
         extra={
             "key_hash": key.hashed_key[:8] + "...",
-            "current_balance": key.balance,
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
+            "current_balance": billing_key.balance,
             "required_cost": cost_per_request,
-            "sufficient_balance": key.balance >= cost_per_request,
+            "sufficient_balance": billing_key.balance >= cost_per_request,
         },
     )
 
-    if key.total_balance < cost_per_request:
+    if billing_key.total_balance < cost_per_request:
         logger.warning(
             "Insufficient balance for request",
             extra={
                 "key_hash": key.hashed_key[:8] + "...",
-                "balance": key.balance,
-                "reserved_balance": key.reserved_balance,
+                "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                "balance": billing_key.balance,
+                "reserved_balance": billing_key.reserved_balance,
                 "required": cost_per_request,
-                "shortfall": cost_per_request - key.total_balance,
+                "shortfall": cost_per_request - billing_key.total_balance,
             },
         )
 
@@ -317,26 +450,78 @@ async def pay_for_request(
             status_code=402,
             detail={
                 "error": {
-                    "message": f"Insufficient balance: {cost_per_request} mSats required. {key.total_balance} available. (reserved: {key.reserved_balance})",
+                    "message": f"Insufficient balance: {cost_per_request} mSats required. {billing_key.total_balance} available. (reserved: {billing_key.reserved_balance})",
                     "type": "insufficient_quota",
                     "code": "insufficient_balance",
                 }
             },
         )
 
+    # Check validity date
+    if key.validity_date is not None:
+        if time.time() > key.validity_date:
+            logger.warning(
+                "Key validity date expired",
+                extra={
+                    "key_hash": key.hashed_key[:8] + "...",
+                    "validity_date": key.validity_date,
+                    "current_time": time.time(),
+                },
+            )
+            raise HTTPException(
+                status_code=403,
+                detail={
+                    "error": {
+                        "message": "API key has expired (validity date reached).",
+                        "type": "invalid_request_error",
+                        "code": "key_expired",
+                    }
+                },
+            )
+
+    # Check balance limit for child keys (or any key with a limit)
+    if key.balance_limit is not None:
+        await check_and_reset_limit(key, session)
+
+        if (
+            key.total_spent + key.reserved_balance + cost_per_request
+            > key.balance_limit
+        ):
+            logger.warning(
+                "Balance limit exceeded",
+                extra={
+                    "key_hash": key.hashed_key[:8] + "...",
+                    "total_spent": key.total_spent,
+                    "reserved": key.reserved_balance,
+                    "balance_limit": key.balance_limit,
+                    "required": cost_per_request,
+                },
+            )
+            raise HTTPException(
+                status_code=402,
+                detail={
+                    "error": {
+                        "message": f"Balance limit exceeded: {key.balance_limit} mSats limit. {key.total_spent} already spent ({key.reserved_balance} reserved), {cost_per_request} required for this request.",
+                        "type": "insufficient_quota",
+                        "code": "balance_limit_exceeded",
+                    }
+                },
+            )
+
     logger.debug(
         "Charging base cost for request",
         extra={
             "key_hash": key.hashed_key[:8] + "...",
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
             "cost": cost_per_request,
-            "balance_before": key.balance,
+            "balance_before": billing_key.balance,
         },
     )
 
     # Charge the base cost for the request atomically to avoid race conditions
     stmt = (
         update(ApiKey)
-        .where(col(ApiKey.hashed_key) == key.hashed_key)
+        .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
         .where(col(ApiKey.balance) - col(ApiKey.reserved_balance) >= cost_per_request)
         .values(
             reserved_balance=col(ApiKey.reserved_balance) + cost_per_request,
@@ -344,6 +529,19 @@ async def pay_for_request(
         )
     )
     result = await session.exec(stmt)  # type: ignore[call-overload]
+
+    # Also increment total_requests and reserved_balance on the child key if it's different
+    if billing_key.hashed_key != key.hashed_key:
+        child_stmt = (
+            update(ApiKey)
+            .where(col(ApiKey.hashed_key) == key.hashed_key)
+            .values(
+                total_requests=col(ApiKey.total_requests) + 1,
+                reserved_balance=col(ApiKey.reserved_balance) + cost_per_request,
+            )
+        )
+        await session.exec(child_stmt)  # type: ignore[call-overload]
+
     await session.commit()
 
     if result.rowcount == 0:
@@ -351,8 +549,9 @@ async def pay_for_request(
             "Concurrent request depleted balance",
             extra={
                 "key_hash": key.hashed_key[:8] + "...",
+                "billing_key_hash": billing_key.hashed_key[:8] + "...",
                 "required_cost": cost_per_request,
-                "current_balance": key.balance,
+                "current_balance": billing_key.balance,
             },
         )
 
@@ -361,23 +560,26 @@ async def pay_for_request(
             status_code=402,
             detail={
                 "error": {
-                    "message": f"Insufficient balance: {cost_per_request} mSats required. {key.balance} available.",
+                    "message": f"Insufficient balance: {cost_per_request} mSats required. {billing_key.balance} available.",
                     "type": "insufficient_quota",
                     "code": "insufficient_balance",
                 }
             },
         )
 
-    await session.refresh(key)
+    await session.refresh(billing_key)
+    if billing_key.hashed_key != key.hashed_key:
+        await session.refresh(key)
 
     logger.info(
         "Payment processed successfully",
         extra={
             "key_hash": key.hashed_key[:8] + "...",
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
             "charged_amount": cost_per_request,
-            "new_balance": key.balance,
-            "total_spent": key.total_spent,
-            "total_requests": key.total_requests,
+            "new_balance": billing_key.balance,
+            "total_spent": billing_key.total_spent,
+            "total_requests": billing_key.total_requests,
         },
     )
 
@@ -387,9 +589,11 @@ async def pay_for_request(
 async def revert_pay_for_request(
     key: ApiKey, session: AsyncSession, cost_per_request: int
 ) -> None:
+    billing_key = await get_billing_key(key, session)
+
     stmt = (
         update(ApiKey)
-        .where(col(ApiKey.hashed_key) == key.hashed_key)
+        .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
         .values(
             reserved_balance=col(ApiKey.reserved_balance) - cost_per_request,
             total_requests=col(ApiKey.total_requests) - 1,
@@ -397,27 +601,43 @@ async def revert_pay_for_request(
     )
 
     result = await session.exec(stmt)  # type: ignore[call-overload]
+
+    # Also decrement total_requests and reserved_balance on the child key if it's different
+    if billing_key.hashed_key != key.hashed_key:
+        child_stmt = (
+            update(ApiKey)
+            .where(col(ApiKey.hashed_key) == key.hashed_key)
+            .values(
+                total_requests=col(ApiKey.total_requests) - 1,
+                reserved_balance=col(ApiKey.reserved_balance) - cost_per_request,
+            )
+        )
+        await session.exec(child_stmt)  # type: ignore[call-overload]
+
     await session.commit()
     if result.rowcount == 0:
         logger.error(
             "Failed to revert payment - insufficient reserved balance",
             extra={
                 "key_hash": key.hashed_key[:8] + "...",
+                "billing_key_hash": billing_key.hashed_key[:8] + "...",
                 "cost_to_revert": cost_per_request,
-                "current_reserved_balance": key.reserved_balance,
+                "current_reserved_balance": billing_key.reserved_balance,
             },
         )
         raise HTTPException(
             status_code=402,
             detail={
                 "error": {
-                    "message": f"failed to revert request payment: {cost_per_request} mSats required. {key.balance} available.",
+                    "message": f"failed to revert request payment: {cost_per_request} mSats required. {billing_key.balance} available.",
                     "type": "payment_error",
                     "code": "payment_error",
                 }
             },
         )
-    await session.refresh(key)
+    await session.refresh(billing_key)
+    if billing_key.hashed_key != key.hashed_key:
+        await session.refresh(key)
 
 
 async def adjust_payment_for_tokens(
@@ -428,25 +648,71 @@ async def adjust_payment_for_tokens(
     This is called after the initial payment and the upstream request is complete.
     Returns cost data to be included in the response.
     """
+    billing_key = await get_billing_key(key, session)
     model = response_data.get("model", "unknown")
 
     logger.debug(
         "Starting payment adjustment for tokens",
         extra={
             "key_hash": key.hashed_key[:8] + "...",
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
             "model": model,
             "deducted_max_cost": deducted_max_cost,
-            "current_balance": key.balance,
+            "current_balance": billing_key.balance,
             "has_usage": "usage" in response_data,
         },
     )
 
+    async def release_reservation_only() -> None:
+        """Fallback to release reservation without charging when main update fails."""
+        try:
+            release_stmt = (
+                update(ApiKey)
+                .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
+                .values(
+                    reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost
+                )
+            )
+            await session.exec(release_stmt)  # type: ignore[call-overload]
+
+            # Also release on child key if it's different
+            if billing_key.hashed_key != key.hashed_key:
+                child_release_stmt = (
+                    update(ApiKey)
+                    .where(col(ApiKey.hashed_key) == key.hashed_key)
+                    .values(
+                        reserved_balance=col(ApiKey.reserved_balance)
+                        - deducted_max_cost
+                    )
+                )
+                await session.exec(child_release_stmt)  # type: ignore[call-overload]
+
+            await session.commit()
+            logger.warning(
+                "Released reservation without charging (fallback)",
+                extra={
+                    "key_hash": key.hashed_key[:8] + "...",
+                    "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                    "deducted_max_cost": deducted_max_cost,
+                },
+            )
+        except Exception as e:
+            logger.error(
+                "Failed to release reservation in fallback",
+                extra={
+                    "error": str(e),
+                    "key_hash": key.hashed_key[:8] + "...",
+                    "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                },
+            )
+
     match await calculate_cost(response_data, deducted_max_cost, session):
         case MaxCostData() as cost:
             logger.debug(
                 "Using max cost data (no token adjustment)",
                 extra={
                     "key_hash": key.hashed_key[:8] + "...",
+                    "billing_key_hash": billing_key.hashed_key[:8] + "...",
                     "model": model,
                     "max_cost": cost.total_msats,
                 },
@@ -454,7 +720,7 @@ async def adjust_payment_for_tokens(
             # Finalize by releasing reservation and charging max cost
             finalize_stmt = (
                 update(ApiKey)
-                .where(col(ApiKey.hashed_key) == key.hashed_key)
+                .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                 .values(
                     reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost,
                     balance=col(ApiKey.balance) - cost.total_msats,
@@ -462,26 +728,45 @@ async def adjust_payment_for_tokens(
                 )
             )
             result = await session.exec(finalize_stmt)  # type: ignore[call-overload]
+
+            # Also update total_spent and reserved_balance on the child key if it's different
+            if billing_key.hashed_key != key.hashed_key:
+                child_stmt = (
+                    update(ApiKey)
+                    .where(col(ApiKey.hashed_key) == key.hashed_key)
+                    .values(
+                        total_spent=col(ApiKey.total_spent) + cost.total_msats,
+                        reserved_balance=col(ApiKey.reserved_balance)
+                        - deducted_max_cost,
+                    )
+                )
+                await session.exec(child_stmt)  # type: ignore[call-overload]
+
             await session.commit()
             if result.rowcount == 0:
                 logger.error(
-                    "Failed to finalize max-cost payment - insufficient reserved balance",
+                    "Failed to finalize max-cost payment - retrying reservation release",
                     extra={
                         "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
                         "deducted_max_cost": deducted_max_cost,
-                        "current_reserved_balance": key.reserved_balance,
+                        "current_reserved_balance": billing_key.reserved_balance,
                         "total_cost": cost.total_msats,
                         "model": model,
                     },
                 )
+                await release_reservation_only()
             else:
-                await session.refresh(key)
+                await session.refresh(billing_key)
+                if billing_key.hashed_key != key.hashed_key:
+                    await session.refresh(key)
                 logger.info(
                     "Max cost payment finalized",
                     extra={
                         "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
                         "charged_amount": cost.total_msats,
-                        "new_balance": key.balance,
+                        "new_balance": billing_key.balance,
                         "model": model,
                     },
                 )
@@ -497,6 +782,7 @@ async def adjust_payment_for_tokens(
                 "Calculated token-based cost",
                 extra={
                     "key_hash": key.hashed_key[:8] + "...",
+                    "billing_key_hash": billing_key.hashed_key[:8] + "...",
                     "model": model,
                     "token_cost": cost.total_msats,
                     "deducted_max_cost": deducted_max_cost,
@@ -509,11 +795,15 @@ async def adjust_payment_for_tokens(
             if cost_difference == 0:
                 logger.debug(
                     "Finalizing with exact reserved cost",
-                    extra={"key_hash": key.hashed_key[:8] + "...", "model": model},
+                    extra={
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                        "model": model,
+                    },
                 )
                 finalize_stmt = (
                     update(ApiKey)
-                    .where(col(ApiKey.hashed_key) == key.hashed_key)
+                    .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                     .values(
                         reserved_balance=col(ApiKey.reserved_balance)
                         - deducted_max_cost,
@@ -522,8 +812,24 @@ async def adjust_payment_for_tokens(
                     )
                 )
                 await session.exec(finalize_stmt)  # type: ignore[call-overload]
+
+                # Also update total_spent and reserved_balance on the child key if it's different
+                if billing_key.hashed_key != key.hashed_key:
+                    child_stmt = (
+                        update(ApiKey)
+                        .where(col(ApiKey.hashed_key) == key.hashed_key)
+                        .values(
+                            total_spent=col(ApiKey.total_spent) + total_cost_msats,
+                            reserved_balance=col(ApiKey.reserved_balance)
+                            - deducted_max_cost,
+                        )
+                    )
+                    await session.exec(child_stmt)  # type: ignore[call-overload]
+
                 await session.commit()
-                await session.refresh(key)
+                await session.refresh(billing_key)
+                if billing_key.hashed_key != key.hashed_key:
+                    await session.refresh(key)
                 return cost.dict()
 
             # this should never happen why do we handle this???
@@ -533,16 +839,17 @@ async def adjust_payment_for_tokens(
                     "Additional charge required for token usage",
                     extra={
                         "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
                         "additional_charge": cost_difference,
-                        "current_balance": key.balance,
-                        "sufficient_balance": key.balance >= cost_difference,
+                        "current_balance": billing_key.balance,
+                        "sufficient_balance": billing_key.balance >= cost_difference,
                         "model": model,
                     },
                 )
 
                 finalize_stmt = (
                     update(ApiKey)
-                    .where(col(ApiKey.hashed_key) == key.hashed_key)
+                    .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                     .values(
                         reserved_balance=col(ApiKey.reserved_balance)
                         - deducted_max_cost,
@@ -551,30 +858,49 @@ async def adjust_payment_for_tokens(
                     )
                 )
                 result = await session.exec(finalize_stmt)  # type: ignore[call-overload]
+
+                # Also update total_spent and reserved_balance on the child key if it's different
+                if billing_key.hashed_key != key.hashed_key:
+                    child_stmt = (
+                        update(ApiKey)
+                        .where(col(ApiKey.hashed_key) == key.hashed_key)
+                        .values(
+                            total_spent=col(ApiKey.total_spent) + total_cost_msats,
+                            reserved_balance=col(ApiKey.reserved_balance)
+                            - deducted_max_cost,
+                        )
+                    )
+                    await session.exec(child_stmt)  # type: ignore[call-overload]
+
                 await session.commit()
 
                 if result.rowcount:
                     cost.total_msats = total_cost_msats
-                    await session.refresh(key)
+                    await session.refresh(billing_key)
+                    if billing_key.hashed_key != key.hashed_key:
+                        await session.refresh(key)
 
                     logger.info(
                         "Finalized payment with additional charge",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
                             "charged_amount": total_cost_msats,
-                            "new_balance": key.balance,
+                            "new_balance": billing_key.balance,
                             "model": model,
                         },
                     )
                 else:
                     logger.warning(
-                        "Failed to finalize additional charge (concurrent operation)",
+                        "Failed to finalize additional charge - releasing reservation",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
                             "attempted_charge": total_cost_msats,
                             "model": model,
                         },
                     )
+                    await release_reservation_only()
             else:
                 # Refund some of the base cost
                 refund = abs(cost_difference)
@@ -582,15 +908,16 @@ async def adjust_payment_for_tokens(
                     "Refunding excess payment",
                     extra={
                         "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
                         "refund_amount": refund,
-                        "current_balance": key.balance,
+                        "current_balance": billing_key.balance,
                         "model": model,
                     },
                 )
 
                 refund_stmt = (
                     update(ApiKey)
-                    .where(col(ApiKey.hashed_key) == key.hashed_key)
+                    .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                     .values(
                         reserved_balance=col(ApiKey.reserved_balance)
                         - deducted_max_cost,
@@ -599,41 +926,58 @@ async def adjust_payment_for_tokens(
                     )
                 )
                 result = await session.exec(refund_stmt)  # type: ignore[call-overload]
+
+                # Also update total_spent and reserved_balance on the child key if it's different
+                if billing_key.hashed_key != key.hashed_key:
+                    child_stmt = (
+                        update(ApiKey)
+                        .where(col(ApiKey.hashed_key) == key.hashed_key)
+                        .values(
+                            total_spent=col(ApiKey.total_spent) + total_cost_msats,
+                            reserved_balance=col(ApiKey.reserved_balance)
+                            - deducted_max_cost,
+                        )
+                    )
+                    await session.exec(child_stmt)  # type: ignore[call-overload]
+
                 await session.commit()
 
                 if result.rowcount == 0:
                     logger.error(
-                        "Failed to finalize payment - insufficient reserved balance",
+                        "Failed to finalize payment - releasing reservation",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
                             "deducted_max_cost": deducted_max_cost,
-                            "current_reserved_balance": key.reserved_balance,
+                            "current_reserved_balance": billing_key.reserved_balance,
                             "total_cost": total_cost_msats,
                             "model": model,
                         },
                     )
-                    # Still return the cost data even if we couldn't properly finalize
-                    # The reservation was already made, so the user has paid
+                    await release_reservation_only()
+                else:
+                    cost.total_msats = total_cost_msats
+                    await session.refresh(billing_key)
+                    if billing_key.hashed_key != key.hashed_key:
+                        await session.refresh(key)
 
-                cost.total_msats = total_cost_msats
-                await session.refresh(key)
-
-                logger.info(
-                    "Refund processed successfully",
-                    extra={
-                        "key_hash": key.hashed_key[:8] + "...",
-                        "refunded_amount": refund,
-                        "new_balance": key.balance,
-                        "final_cost": cost.total_msats,
-                        "model": model,
-                    },
-                )
+                    logger.info(
+                        "Refund processed successfully",
+                        extra={
+                            "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                            "refunded_amount": refund,
+                            "new_balance": billing_key.balance,
+                            "final_cost": cost.total_msats,
+                            "model": model,
+                        },
+                    )
 
             return cost.dict()
 
         case CostDataError() as error:
             logger.error(
-                "Cost calculation error during payment adjustment",
+                "Cost calculation error during payment adjustment - releasing reservation",
                 extra={
                     "key_hash": key.hashed_key[:8] + "...",
                     "model": model,
@@ -641,6 +985,7 @@ async def adjust_payment_for_tokens(
                     "error_code": error.code,
                 },
             )
+            await release_reservation_only()
 
             raise HTTPException(
                 status_code=400,
@@ -652,10 +997,77 @@ async def adjust_payment_for_tokens(
                     }
                 },
             )
-    # Fallback return to satisfy type checker; execution should not reach here
+    # Fallback: should not reach here, but release reservation just in case
+    logger.error(
+        "Unexpected fallback in adjust_payment_for_tokens - releasing reservation",
+        extra={"key_hash": key.hashed_key[:8] + "...", "model": model},
+    )
+    await release_reservation_only()
     return {
         "base_msats": deducted_max_cost,
         "input_msats": 0,
         "output_msats": 0,
         "total_msats": deducted_max_cost,
     }
+
+
+async def periodic_key_reset() -> None:
+    """Background task to reset key limits based on their policy."""
+    from .core.db import create_session
+
+    while True:
+        try:
+            interval = 3600  # Run every hour
+            jitter = 300
+            await asyncio.sleep(interval + random.uniform(0, jitter))
+        except asyncio.CancelledError:
+            break
+
+        try:
+            async with create_session() as session:
+                # Find all keys that have a reset policy
+                stmt = select(ApiKey).where(ApiKey.balance_limit_reset.is_not(None))  # type: ignore
+                keys = (await session.exec(stmt)).all()
+
+                now = int(time.time())
+                updated_count = 0
+
+                for key in keys:
+                    reset_date = key.balance_limit_reset_date or 0
+                    should_reset = False
+
+                    if key.balance_limit_reset == "daily":
+                        if (
+                            datetime.fromtimestamp(now).date()
+                            > datetime.fromtimestamp(reset_date).date()
+                        ):
+                            should_reset = True
+                    elif key.balance_limit_reset == "weekly":
+                        if (
+                            datetime.fromtimestamp(now).isocalendar()[:2]
+                            > datetime.fromtimestamp(reset_date).isocalendar()[:2]
+                        ):
+                            should_reset = True
+                    elif key.balance_limit_reset == "monthly":
+                        dt_now = datetime.fromtimestamp(now)
+                        dt_reset = datetime.fromtimestamp(reset_date)
+                        if dt_now.year > dt_reset.year or dt_now.month > dt_reset.month:
+                            should_reset = True
+
+                    if should_reset:
+                        key.total_spent = 0
+                        key.balance_limit_reset_date = now
+                        session.add(key)
+                        updated_count += 1
+
+                if updated_count > 0:
+                    await session.commit()
+                    logger.info(
+                        "Periodic key reset complete",
+                        extra={"keys_reset": updated_count},
+                    )
+
+        except asyncio.CancelledError:
+            break
+        except Exception as e:
+            logger.error(f"Error in periodic_key_reset: {e}")

routstr/balance.py

@@ -1,5 +1,6 @@
 import asyncio
 import hashlib
+import time
 from time import monotonic
 from typing import Annotated, NoReturn
 
@@ -32,16 +33,33 @@ async def get_key_from_header(
     )
 
 
-# TODO: remove this endpoint when frontend is updated
-@router.get("/", include_in_schema=False)
-async def account_info(key: ApiKey = Depends(get_key_from_header)) -> dict:
+async def get_balance_info(key: ApiKey, session: AsyncSession) -> dict:
+    from .auth import get_billing_key
+
+    billing_key = await get_billing_key(key, session)
     return {
         "api_key": "sk-" + key.hashed_key,
-        "balance": key.balance,
-        "reserved": key.reserved_balance,
+        "balance": billing_key.balance,
+        "reserved": billing_key.reserved_balance,
+        "is_child": key.parent_key_hash is not None,
+        "parent_key": "sk-" + key.parent_key_hash if key.parent_key_hash else None,
+        "total_requests": key.total_requests,
+        "total_spent": key.total_spent,
+        "balance_limit": key.balance_limit,
+        "balance_limit_reset": key.balance_limit_reset,
+        "validity_date": key.validity_date,
     }
 
 
+# TODO: remove this endpoint when frontend is updated
+@router.get("/", include_in_schema=False)
+async def account_info(
+    key: ApiKey = Depends(get_key_from_header),
+    session: AsyncSession = Depends(get_session),
+) -> dict:
+    return await get_balance_info(key, session)
+
+
 # TODO: Implement POST /v1/wallet/create endpoint
 # This endpoint should accept:
 # - cashu_token (required): The eCash token to deposit
@@ -56,9 +74,24 @@ async def account_info(key: ApiKey = Depends(get_key_from_header)) -> dict:
 
 @router.get("/create")
 async def create_balance(
-    initial_balance_token: str, session: AsyncSession = Depends(get_session)
+    initial_balance_token: str,
+    balance_limit: int | None = None,
+    balance_limit_reset: str | None = None,
+    validity_date: int | None = None,
+    session: AsyncSession = Depends(get_session),
 ) -> dict:
     key = await validate_bearer_key(initial_balance_token, session)
+
+    if balance_limit is not None or balance_limit_reset or validity_date:
+        key.balance_limit = balance_limit
+        key.balance_limit_reset = balance_limit_reset
+        key.validity_date = validity_date
+        if balance_limit_reset:
+            key.balance_limit_reset_date = int(time.time())
+        session.add(key)
+        await session.commit()
+        await session.refresh(key)
+
     return {
         "api_key": "sk-" + key.hashed_key,
         "balance": key.balance,
@@ -66,12 +99,11 @@ async def create_balance(
 
 
 @router.get("/info")
-async def wallet_info(key: ApiKey = Depends(get_key_from_header)) -> dict:
-    return {
-        "api_key": "sk-" + key.hashed_key,
-        "balance": key.balance,
-        "reserved": key.reserved_balance,
-    }
+async def wallet_info(
+    key: ApiKey = Depends(get_key_from_header),
+    session: AsyncSession = Depends(get_session),
+) -> dict:
+    return await get_balance_info(key, session)
 
 
 class TopupRequest(BaseModel):
@@ -85,6 +117,10 @@ async def topup_wallet_endpoint(
     key: ApiKey = Depends(get_key_from_header),
     session: AsyncSession = Depends(get_session),
 ) -> dict[str, int]:
+    from .auth import get_billing_key
+
+    billing_key = await get_billing_key(key, session)
+
     if topup_request is not None:
         cashu_token = topup_request.cashu_token
     if cashu_token is None:
@@ -94,7 +130,7 @@ async def topup_wallet_endpoint(
     if len(cashu_token) < 10 or "cashu" not in cashu_token:
         raise HTTPException(status_code=400, detail="Invalid token format")
     try:
-        amount_msats = await credit_balance(cashu_token, key, session)
+        amount_msats = await credit_balance(cashu_token, billing_key, session)
     except ValueError as e:
         error_msg = str(e)
         if "already spent" in error_msg.lower():
@@ -150,11 +186,18 @@ async def refund_wallet_endpoint(
 
     bearer_value: str = authorization[7:]
 
+    key: ApiKey = await validate_bearer_key(bearer_value, session)
+
     if cached := await _refund_cache_get(bearer_value):
         return cached
 
-    key: ApiKey = await validate_bearer_key(bearer_value, session)
-    remaining_balance_msats: int = key.balance
+    if key.parent_key_hash:
+        raise HTTPException(
+            status_code=400,
+            detail="Cannot refund child key. Please refund the parent key instead.",
+        )
+
+    remaining_balance_msats: int = key.total_balance
 
     if key.refund_currency == "sat":
         remaining_balance = remaining_balance_msats // 1000
@@ -208,7 +251,9 @@ async def refund_wallet_endpoint(
 
     await _refund_cache_set(bearer_value, result)
 
-    await session.delete(key)
+    key.balance = 0
+    key.reserved_balance = 0
+    session.add(key)
     await session.commit()
 
     return result
@@ -227,6 +272,117 @@ async def donate(token: str, ref: str | None = None) -> str:
         return "Invalid token."
 
 
+class ChildKeyRequest(BaseModel):
+    count: int
+    balance_limit: int | None = None
+    balance_limit_reset: str | None = None
+    validity_date: int | None = None
+
+
+@router.post("/child-key")
+async def create_child_key(
+    payload: ChildKeyRequest,
+    key: ApiKey = Depends(get_key_from_header),
+    session: AsyncSession = Depends(get_session),
+) -> dict:
+    """Creates one or more child API keys that use the parent's balance."""
+    # Log incoming request for debugging
+    logger.debug(f"Child key creation request: count={payload.count}")
+
+    count = payload.count
+    if count < 1 or count > 50:
+        raise HTTPException(status_code=400, detail="Count must be between 1 and 50.")
+
+    # Check if this is already a child key
+    if key.parent_key_hash:
+        raise HTTPException(
+            status_code=400,
+            detail="Cannot create a child key for another child key.",
+        )
+
+    cost_per_key = settings.child_key_cost
+    total_cost = cost_per_key * count
+
+    if key.total_balance < total_cost:
+        raise HTTPException(
+            status_code=402,
+            detail=f"Insufficient balance to create {count} child keys. {total_cost} mSats required.",
+        )
+
+    # Deduct cost from parent
+    key.balance -= total_cost
+    key.total_spent += total_cost
+    session.add(key)
+
+    # Generate new keys
+    import secrets
+
+    new_keys = []
+    for _ in range(count):
+        new_key_raw = secrets.token_hex(32)
+        new_key_hash = new_key_raw  # We use the raw key as the hash for sk- keys
+
+        child_key = ApiKey(
+            hashed_key=new_key_hash,
+            balance=0,
+            parent_key_hash=key.hashed_key,
+            balance_limit=payload.balance_limit,
+            balance_limit_reset=payload.balance_limit_reset,
+            balance_limit_reset_date=int(time.time())
+            if payload.balance_limit_reset
+            else None,
+            validity_date=payload.validity_date,
+        )
+        session.add(child_key)
+        new_keys.append("sk-" + new_key_hash)
+
+    await session.commit()
+
+    response_data = {
+        "api_keys": new_keys,
+        "count": count,
+        "cost_msats": total_cost,
+        "cost_sats": total_cost // 1000,
+        "parent_balance": key.balance,
+        "parent_balance_sats": key.balance // 1000,
+    }
+    logger.debug(f"Child key creation response: {response_data}")
+    return response_data
+
+
+class ChildKeyResetRequest(BaseModel):
+    child_key: str
+
+
+@router.post("/child-key/reset")
+async def reset_child_key_spent(
+    payload: ChildKeyResetRequest,
+    key: ApiKey = Depends(get_key_from_header),
+    session: AsyncSession = Depends(get_session),
+) -> dict:
+    """Resets the total_spent of a child key. Must be called by the parent."""
+    child_key_raw = payload.child_key
+    if child_key_raw.startswith("sk-"):
+        child_key_raw = child_key_raw[3:]
+
+    child_key = await session.get(ApiKey, child_key_raw)
+    if not child_key:
+        raise HTTPException(status_code=404, detail="Child key not found.")
+
+    if child_key.parent_key_hash != key.hashed_key:
+        raise HTTPException(
+            status_code=403, detail="Unauthorized. You are not the parent of this key."
+        )
+
+    child_key.total_spent = 0
+    if child_key.balance_limit_reset:
+        child_key.balance_limit_reset_date = int(time.time())
+    session.add(child_key)
+    await session.commit()
+
+    return {"success": True, "message": "Child key balance reset successfully."}
+
+
 @router.api_route(
     "/{path:path}",
     methods=["GET", "POST", "PUT", "DELETE"],

routstr/core/db.py

@@ -5,8 +5,9 @@ from typing import AsyncGenerator
 
 from alembic import command
 from alembic.config import Config
+from sqlalchemy import UniqueConstraint
 from sqlalchemy.ext.asyncio.engine import create_async_engine
-from sqlmodel import Field, Relationship, SQLModel, func, select
+from sqlmodel import Field, Relationship, SQLModel, func, select, update
 from sqlmodel.ext.asyncio.session import AsyncSession
 
 from .logging import get_logger
@@ -47,12 +48,39 @@ class ApiKey(SQLModel, table=True):  # type: ignore
         default=None,
         description="Currency of the cashu-token",
     )
+    parent_key_hash: str | None = Field(
+        default=None, foreign_key="api_keys.hashed_key", index=True
+    )
+    balance_limit: int | None = Field(
+        default=None,
+        description="Max spendable balance in msats for this key (mostly for child keys)",
+    )
+    balance_limit_reset: str | None = Field(
+        default=None,
+        description="Reset policy for balance limit (manual, daily, monthly, etc.)",
+    )
+    balance_limit_reset_date: int | None = Field(
+        default=None,
+        description="Unix timestamp of the last time the balance limit was reset",
+    )
+    validity_date: int | None = Field(
+        default=None,
+        description="Unix timestamp after which the key is no longer valid",
+    )
 
     @property
     def total_balance(self) -> int:
         return self.balance - self.reserved_balance
 
 
+async def reset_all_reserved_balances(session: AsyncSession) -> None:
+    logger.info("Resetting all reserved balances to 0")
+    stmt = update(ApiKey).values(reserved_balance=0)
+    await session.exec(stmt)  # type: ignore[call-overload]
+    await session.commit()
+    logger.info("Reserved balances reset successfully")
+
+
 class ModelRow(SQLModel, table=True):  # type: ignore
     __tablename__ = "models"
     id: str = Field(primary_key=True)
@@ -100,11 +128,16 @@ class LightningInvoice(SQLModel, table=True):  # type: ignore
 
 class UpstreamProviderRow(SQLModel, table=True):  # type: ignore
     __tablename__ = "upstream_providers"
+    __table_args__ = (
+        UniqueConstraint(
+            "base_url", "api_key", name="uq_upstream_providers_base_url_api_key"
+        ),
+    )
     id: int | None = Field(default=None, primary_key=True)
     provider_type: str = Field(
         description="Provider type: custom, openai, anthropic, azure, openrouter, etc."
     )
-    base_url: str = Field(unique=True, description="Base URL of the upstream API")
+    base_url: str = Field(description="Base URL of the upstream API")
     api_key: str = Field(description="API key for the upstream provider")
     api_version: str | None = Field(
         default=None, description="API version for Azure OpenAI"

routstr/core/exceptions.py

@@ -6,6 +6,15 @@ from .logging import get_logger
 logger = get_logger(__name__)
 
 
+class UpstreamError(Exception):
+    """Exception raised when an upstream provider fails."""
+
+    def __init__(self, message: str, status_code: int = 502):
+        self.message = message
+        self.status_code = status_code
+        super().__init__(message)
+
+
 async def http_exception_handler(request: Request, exc: Exception) -> JSONResponse:
     """Handle HTTP exceptions and include request ID in response."""
     request_id = getattr(request.state, "request_id", "unknown")

routstr/core/log_manager.py

@@ -105,6 +105,9 @@ class LogManager:
         level: str | None = None,
         request_id: str | None = None,
         search_text: str | None = None,
+        status_codes: list[int] | None = None,
+        methods: list[str] | None = None,
+        endpoints: list[str] | None = None,
         limit: int = 100,
     ) -> list[dict[str, Any]]:
         """
@@ -134,7 +137,13 @@ class LogManager:
 
         for log_data in iterator:
             if not self._matches_filters(
-                log_data, level, request_id, search_text_lower
+                log_data,
+                level,
+                request_id,
+                search_text_lower,
+                status_codes,
+                methods,
+                endpoints,
             ):
                 continue
 
@@ -153,6 +162,9 @@ class LogManager:
         level: str | None,
         request_id: str | None,
         search_text_lower: str | None,
+        status_codes: list[int] | None = None,
+        methods: list[str] | None = None,
+        endpoints: list[str] | None = None,
     ) -> bool:
         if level and log_data.get("levelname", "").upper() != level.upper():
             return False
@@ -160,6 +172,36 @@ class LogManager:
         if request_id and log_data.get("request_id") != request_id:
             return False
 
+        if status_codes:
+            entry_status = log_data.get("status_code")
+            if entry_status is not None:
+                try:
+                    if int(entry_status) not in status_codes:
+                        return False
+                except (ValueError, TypeError):
+                    return False
+            else:
+                return False
+
+        if methods:
+            entry_method = log_data.get("method", "").upper()
+            if entry_method not in [m.upper() for m in methods]:
+                return False
+
+        if endpoints:
+            entry_path = log_data.get("path", "")
+            matched = False
+            for endpoint in endpoints:
+                clean_endpoint = endpoint.lstrip("/")
+                if entry_path.startswith(clean_endpoint):
+                    matched = True
+                    break
+                if clean_endpoint in entry_path:
+                    matched = True
+                    break
+            if not matched:
+                return False
+
         if search_text_lower:
             message = str(log_data.get("message", "")).lower()
             name = str(log_data.get("name", "")).lower()

routstr/core/main.py

@@ -10,13 +10,11 @@ from fastapi.responses import FileResponse, RedirectResponse
 from fastapi.staticfiles import StaticFiles
 from starlette.exceptions import HTTPException
 
+from ..auth import periodic_key_reset
 from ..balance import balance_router, deprecated_wallet_router
-from ..discovery import providers_cache_refresher, providers_router
-from ..nip91 import announce_provider
-from ..payment.models import (
-    models_router,
-    update_sats_pricing,
-)
+from ..nostr import announce_provider, providers_cache_refresher
+from ..nostr.discovery import providers_router
+from ..payment.models import models_router, update_sats_pricing
 from ..payment.price import update_prices_periodically
 from ..proxy import initialize_upstreams, proxy_router, refresh_model_maps_periodically
 from ..wallet import periodic_payout
@@ -33,9 +31,9 @@ setup_logging()
 logger = get_logger(__name__)
 
 if os.getenv("VERSION_SUFFIX") is not None:
-    __version__ = f"0.2.1-{os.getenv('VERSION_SUFFIX')}"
+    __version__ = f"0.3.0-{os.getenv('VERSION_SUFFIX')}"
 else:
-    __version__ = "0.2.1"
+    __version__ = "0.3.0"
 
 
 @asynccontextmanager
@@ -49,6 +47,7 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
     providers_task = None
     models_refresh_task = None
     model_maps_refresh_task = None
+    key_reset_task = None
 
     try:
         # Run database migrations on startup
@@ -61,6 +60,15 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
         # Initialize application settings (env -> computed -> DB precedence)
         async with create_session() as session:
             s = await SettingsService.initialize(session)
+            if s.reset_reserved_balance_on_startup:
+                from .db import reset_all_reserved_balances
+
+                await reset_all_reserved_balances(session)
+
+        if not s.admin_password:
+            logger.warning(
+                f"Admin password is not set. Visit {s.http_url or 'http://localhost:8000'}/admin to set the password."
+            )
 
         # Apply app metadata from settings
         try:
@@ -95,6 +103,7 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
             nip91_task = asyncio.create_task(announce_provider())
         if global_settings.providers_refresh_interval_seconds > 0:
             providers_task = asyncio.create_task(providers_cache_refresher())
+        key_reset_task = asyncio.create_task(periodic_key_reset())
 
         yield
 
@@ -124,6 +133,8 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
             models_refresh_task.cancel()
         if model_maps_refresh_task is not None:
             model_maps_refresh_task.cancel()
+        if key_reset_task is not None:
+            key_reset_task.cancel()
 
         try:
             tasks_to_wait = []
@@ -141,6 +152,8 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
                 tasks_to_wait.append(models_refresh_task)
             if model_maps_refresh_task is not None:
                 tasks_to_wait.append(model_maps_refresh_task)
+            if key_reset_task is not None:
+                tasks_to_wait.append(key_reset_task)
 
             if tasks_to_wait:
                 await asyncio.gather(*tasks_to_wait, return_exceptions=True)
@@ -182,6 +195,7 @@ async def info() -> dict:
         "mints": global_settings.cashu_mints,
         "http_url": global_settings.http_url,
         "onion_url": global_settings.onion_url,
+        "child_key_cost_msats": global_settings.child_key_cost,
     }
 
 

routstr/core/settings.py

@@ -6,7 +6,7 @@ import os
 from datetime import datetime, timezone
 from typing import Any
 
-from pydantic.v1 import BaseModel, BaseSettings, Field, validator
+from pydantic.v1 import BaseModel, BaseSettings, Field
 from sqlmodel.ext.asyncio.session import AsyncSession
 
 
@@ -37,13 +37,6 @@ class Settings(BaseSettings):
 
     # Cashu
     cashu_mints: list[str] = Field(default_factory=list, env="CASHU_MINTS")
-
-    @validator("cashu_mints", pre=True, each_item=True)
-    def normalize_mint_url(cls, v: str) -> str:
-        if isinstance(v, str):
-            return v.rstrip("/")
-        return v
-
     receive_ln_address: str = Field(default="", env="RECEIVE_LN_ADDRESS")
     primary_mint: str = Field(default="", env="PRIMARY_MINT_URL")
     primary_mint_unit: str = Field(default="sat", env="PRIMARY_MINT_UNIT")
@@ -59,14 +52,18 @@ class Settings(BaseSettings):
     exchange_fee: float = Field(default=1.005, env="EXCHANGE_FEE")
     upstream_provider_fee: float = Field(default=1.05, env="UPSTREAM_PROVIDER_FEE")
     tolerance_percentage: float = Field(default=1.0, env="TOLERANCE_PERCENTAGE")
+    child_key_cost: int = Field(default=1000, env="CHILD_KEY_COST")
     # Minimum per-request charge in millisatoshis when model pricing is free/zero
     min_request_msat: int = Field(default=1, env="MIN_REQUEST_MSAT")
+    reset_reserved_balance_on_startup: bool = Field(
+        default=True, env="RESET_RESERVED_BALANCE_ON_STARTUP"
+    )  # deactivate in horizontal scaling setups
 
     # Network
     cors_origins: list[str] = Field(default_factory=lambda: ["*"], env="CORS_ORIGINS")
     tor_proxy_url: str = Field(default="socks5://127.0.0.1:9050", env="TOR_PROXY_URL")
     providers_refresh_interval_seconds: int = Field(
-        default=300, env="PROVIDERS_REFRESH_INTERVAL_SECONDS"
+        default=0, env="PROVIDERS_REFRESH_INTERVAL_SECONDS"
     )
     pricing_refresh_interval_seconds: int = Field(
         default=120, env="PRICING_REFRESH_INTERVAL_SECONDS"
@@ -146,7 +143,7 @@ def resolve_bootstrap() -> Settings:
         pass
     if not base.onion_url:
         try:
-            from ..nip91 import discover_onion_url_from_tor  # type: ignore
+            from ..nostr.listing import discover_onion_url_from_tor  # type: ignore
 
             discovered = discover_onion_url_from_tor()
             if discovered:

routstr/lightning.py

@@ -23,6 +23,9 @@ class InvoiceCreateRequest(BaseModel):
     api_key: str | None = Field(
         default=None, description="Required for topup operations"
     )
+    balance_limit: int | None = Field(default=None)
+    balance_limit_reset: str | None = Field(default=None)
+    validity_date: int | None = Field(default=None)
 
 
 class InvoiceCreateResponse(BaseModel):
@@ -94,6 +97,9 @@ async def create_invoice(
             status="pending",
             api_key_hash=request.api_key[3:] if request.api_key else None,
             purpose=request.purpose,
+            balance_limit=request.balance_limit,
+            balance_limit_reset=request.balance_limit_reset,
+            validity_date=request.validity_date,
             expires_at=expires_at,
         )
 

routstr/nostr/__init__.py

@@ -0,0 +1,4 @@
+from .discovery import providers_cache_refresher
+from .listing import announce_provider
+
+__all__ = ["providers_cache_refresher", "announce_provider"]

routstr/nostr/discovery.py

@@ -6,10 +6,10 @@ from typing import Any
 
 import httpx
 import websockets
-from fastapi import APIRouter
+from fastapi import APIRouter, HTTPException
 
-from .core.logging import get_logger
-from .core.settings import settings
+from ..core.logging import get_logger
+from ..core.settings import settings
 
 logger = get_logger(__name__)
 
@@ -72,8 +72,6 @@ async def query_nostr_relay_for_providers(
                     elif data[0] == "NOTICE":
                         try:
                             msg = str(data[1])
-                            if len(msg) > 200:
-                                msg = msg[:200] + "..."
                             logger.debug(f"Relay notice: {msg}")
                         except Exception:
                             logger.debug("Relay notice received")
@@ -389,6 +387,9 @@ async def get_providers(
     Return cached providers. If include_json, return provider+health; otherwise provider only.
     Optional filter by pubkey.
     """
+    if settings.providers_refresh_interval_seconds == 0:
+        raise HTTPException(status_code=404, detail="Provider discovery is disabled")
+
     cache = await get_cache()
     if not cache:
         await refresh_providers_cache(pubkey=pubkey)

routstr/nostr/listing.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 """
-NIP-91: Routstr Provider Discoverability Implementation
+Listing: Routstr Provider Discoverability Implementation
 Automatically announces this Routstr proxy instance to Nostr relays.
 """
 
@@ -18,15 +18,15 @@ from nostr.key import PrivateKey
 from nostr.message_type import ClientMessageType
 from nostr.relay_manager import RelayManager
 
-from .core import get_logger
-from .core.settings import settings
+from ..core import get_logger
+from ..core.settings import settings
 
 logger = get_logger(__name__)
 
 
 def get_app_version() -> str | None:
     try:
-        from .core.main import __version__ as imported_version
+        from ..core.main import __version__ as imported_version
 
         return imported_version
     except Exception:
@@ -71,7 +71,7 @@ def nsec_to_keypair(nsec: str) -> tuple[str, str] | None:
         return None
 
 
-def create_nip91_event(
+def create_listing_event(
     private_key_hex: str,
     provider_id: str,
     endpoint_urls: list[str],
@@ -80,7 +80,7 @@ def create_nip91_event(
     metadata: dict[str, Any] | None = None,
 ) -> dict[str, Any]:
     """
-    Create a NIP-91 compliant provider announcement event (kind:38421).
+    Create a listing provider announcement event (kind:38421).
 
     Args:
         private_key_hex: 32-byte hex private key for signing
@@ -164,14 +164,14 @@ def events_semantically_equal(a: dict[str, Any], b: dict[str, Any]) -> bool:
     return True
 
 
-async def query_nip91_events(
+async def query_listing_events(
     relay_url: str,
     pubkey: str,
     provider_id: str | None = None,
     timeout: int = 30,
 ) -> tuple[list[dict[str, Any]], bool]:
     """
-    Query a Nostr relay for NIP-91 provider announcements (kind:38421) via nostr library.
+    Query a Nostr relay for listing provider announcements (kind:38421) via nostr library.
 
     Returns a tuple of (events, ok) where ok indicates whether the relay interaction
     succeeded without transport-level errors.
@@ -188,7 +188,7 @@ async def query_nip91_events(
 
             flt = Filter(kinds=[38421], authors=[pubkey], limit=10)
             filters = Filters([flt])
-            sub_id = f"nip91_{int(time.time())}"
+            sub_id = f"routstr_listing_{int(time.time())}"
             rm.add_subscription(sub_id, filters)
             req: list[Any] = [ClientMessageType.REQUEST, sub_id]
             req.extend(filters.to_json_array())
@@ -294,7 +294,7 @@ async def _determine_provider_id(public_key_hex: str, relay_urls: list[str]) ->
 
     async def query_single_relay(relay_url: str) -> list[dict[str, Any]]:
         try:
-            events, _ok = await query_nip91_events(relay_url, public_key_hex, None)
+            events, _ok = await query_listing_events(relay_url, public_key_hex, None)
             return events
         except Exception:
             return []
@@ -330,7 +330,7 @@ async def publish_to_relay(
     timeout: int = 30,
 ) -> bool:
     """
-    Publish a NIP-91 event to a nostr relay via nostr library.
+    Publish a listing event to a nostr relay via nostr library.
     """
 
     def _sync_publish() -> bool:
@@ -341,7 +341,7 @@ async def publish_to_relay(
             time.sleep(1.0)
             # Publish the event as-is via publish_message to preserve signature
             rm.publish_message(json.dumps(["EVENT", event]))
-            logger.debug(f"Sent NIP-91 event {event.get('id', '')} to {relay_url}")
+            logger.debug(f"Sent listing event {event.get('id', '')} to {relay_url}")
             time.sleep(1.0)
             return True
         except Exception as e:
@@ -364,13 +364,13 @@ async def announce_provider() -> None:
     # Check for NSEC in environment (use NSEC only)
     nsec = settings.nsec
     if not nsec:
-        logger.info("Nostr private key not found (NSEC), skipping NIP-91 announcement")
+        logger.info("Nostr private key not found (NSEC), skipping listing announcement")
         return
 
     # Convert NSEC to keypair
     keypair = nsec_to_keypair(nsec)
     if not keypair:
-        logger.error("Failed to parse NSEC, skipping NIP-91 announcement")
+        logger.error("Failed to parse NSEC, skipping listing announcement")
         return
 
     private_key_hex, public_key_hex = keypair
@@ -409,7 +409,7 @@ async def announce_provider() -> None:
 
     if not endpoint_urls:
         logger.warning(
-            "No valid endpoints configured (HTTP_URL/ONION_URL). Skipping NIP-91 publish."
+            "No valid endpoints configured (HTTP_URL/ONION_URL). Skipping listing publish."
         )
         return
 
@@ -434,7 +434,7 @@ async def announce_provider() -> None:
 
     # Create the candidate event that we would publish
     version_str = get_app_version()
-    candidate_event = create_nip91_event(
+    candidate_event = create_listing_event(
         private_key_hex=private_key_hex,
         provider_id=provider_id,
         endpoint_urls=endpoint_urls,
@@ -474,7 +474,7 @@ async def announce_provider() -> None:
         if _should_skip(relay_url):
             logger.debug(f"Skipping {relay_url} due to backoff")
             continue
-        events, ok = await query_nip91_events(relay_url, public_key_hex, provider_id)
+        events, ok = await query_listing_events(relay_url, public_key_hex, provider_id)
         if ok:
             _register_success(relay_url)
             existing_events.extend(events)
@@ -489,7 +489,7 @@ async def announce_provider() -> None:
 
     if not all_match:
         logger.debug(
-            "No matching NIP-91 announcement found or differences detected; publishing update"
+            "No matching listing announcement found or differences detected; publishing update"
         )
         success_count = 0
         for relay_url in relay_urls:
@@ -502,11 +502,11 @@ async def announce_provider() -> None:
             else:
                 _register_failure(relay_url)
         logger.info(
-            f"Published NIP-91 announcement to {success_count}/{len(relay_urls)} relays"
+            f"Published listing announcement to {success_count}/{len(relay_urls)} relays"
         )
     else:
         logger.debug(
-            "Matching NIP-91 announcement already present; skipping publish on startup"
+            "Matching listing announcement already present; skipping publish on startup"
         )
 
     # Re-announce periodically (every 24 hours)
@@ -518,7 +518,7 @@ async def announce_provider() -> None:
 
             # Build fresh candidate event for comparison
             version_str = get_app_version()
-            candidate_event = create_nip91_event(
+            candidate_event = create_listing_event(
                 private_key_hex=private_key_hex,
                 provider_id=provider_id,
                 endpoint_urls=endpoint_urls,
@@ -533,7 +533,7 @@ async def announce_provider() -> None:
                 if _should_skip(relay_url):
                     logger.debug(f"Skipping {relay_url} due to backoff")
                     continue
-                events, ok = await query_nip91_events(
+                events, ok = await query_listing_events(
                     relay_url, public_key_hex, provider_id
                 )
                 if ok:
@@ -549,7 +549,7 @@ async def announce_provider() -> None:
 
             if all_match:
                 logger.debug(
-                    "Matching NIP-91 announcement already present; skipping periodic re-announce"
+                    "Matching listing announcement already present; skipping periodic re-announce"
                 )
                 continue
 
@@ -567,8 +567,8 @@ async def announce_provider() -> None:
                     _register_failure(relay_url)
 
         except asyncio.CancelledError:
-            logger.info("NIP-91 announcement task cancelled")
+            logger.info("Listing announcement task cancelled")
             break
         except Exception as e:
-            logger.debug(f"Error in NIP-91 announcement loop: {type(e).__name__}")
+            logger.debug(f"Error in listing announcement loop: {type(e).__name__}")
             # Continue running despite errors

routstr/payment/cost_calculation.py

@@ -15,6 +15,7 @@ class CostData(BaseModel):
     input_msats: int
     output_msats: int
     total_msats: int
+    total_usd: float = 0.0
 
 
 class MaxCostData(CostData):
@@ -48,13 +49,6 @@ async def calculate_cost(  # todo: can be sync
         },
     )
 
-    cost_data = MaxCostData(
-        base_msats=max_cost,
-        input_msats=0,
-        output_msats=0,
-        total_msats=max_cost,
-    )
-
     if "usage" not in response_data or response_data["usage"] is None:
         logger.warning(
             "No usage data in response, using base cost only",
@@ -63,7 +57,13 @@ async def calculate_cost(  # todo: can be sync
                 "model": response_data.get("model", "unknown"),
             },
         )
-        return cost_data
+        return MaxCostData(
+            base_msats=0,
+            input_msats=0,
+            output_msats=0,
+            total_msats=0,
+            total_usd=0.0,
+        )
 
     usage_data = response_data["usage"]
 
@@ -103,6 +103,7 @@ async def calculate_cost(  # todo: can be sync
                 input_msats=-1,  # Cost field doesn't break down by token type
                 output_msats=-1,
                 total_msats=cost_in_msats,
+                total_usd=usd_cost,
             )
         except Exception as e:
             logger.warning(
@@ -178,7 +179,12 @@ async def calculate_cost(  # todo: can be sync
                 "model": response_data.get("model", "unknown"),
             },
         )
-        return cost_data
+        return MaxCostData(
+            base_msats=max_cost,
+            input_msats=0,
+            output_msats=0,
+            total_msats=max_cost,
+        )
 
     input_tokens = usage_data.get("prompt_tokens", 0)
     output_tokens = usage_data.get("completion_tokens", 0)
@@ -192,13 +198,22 @@ async def calculate_cost(  # todo: can be sync
     )
 
     # added for response api
-    input_tokens = input_tokens if input_tokens != 0 else  response_data.get("usage", {}).get("input_tokens", 0)
-    output_tokens = output_tokens if output_tokens != 0 else  response_data.get("usage", {}).get("output_tokens", 0)
+    input_tokens = (
+        input_tokens
+        if input_tokens != 0
+        else response_data.get("usage", {}).get("input_tokens", 0)
+    )
+    output_tokens = (
+        output_tokens
+        if output_tokens != 0
+        else response_data.get("usage", {}).get("output_tokens", 0)
+    )
 
     input_msats = round(input_tokens / 1000 * MSATS_PER_1K_INPUT_TOKENS, 3)
 
     output_msats = round(output_tokens / 1000 * MSATS_PER_1K_OUTPUT_TOKENS, 3)
     token_based_cost = math.ceil(input_msats + output_msats)
+    total_usd = (token_based_cost / 1000.0) * sats_usd_price()
 
     logger.info(
         "Calculated token-based cost",
@@ -208,6 +223,7 @@ async def calculate_cost(  # todo: can be sync
             "input_cost_msats": input_msats,
             "output_cost_msats": output_msats,
             "total_cost_msats": token_based_cost,
+            "total_usd": total_usd,
             "model": response_data.get("model", "unknown"),
         },
     )
@@ -217,4 +233,5 @@ async def calculate_cost(  # todo: can be sync
         input_msats=int(input_msats),
         output_msats=int(output_msats),
         total_msats=token_based_cost,
+        total_usd=total_usd,
     )

routstr/payment/lnurl.py

@@ -25,82 +25,6 @@ class LNURLError(Exception):
     """LNURL related errors."""
 
 
-def parse_lightning_invoice_amount(invoice: str, currency: str = "sat") -> int:
-    """Parse Lightning invoice (BOLT-11) to extract amount in specified currency units.
-
-    Args:
-        invoice: BOLT-11 Lightning invoice string
-        currency: Target currency unit ("sat" or "msat")
-
-    Returns:
-        Amount in the specified currency unit
-
-    Raises:
-        LNURLError: If invoice format is invalid or amount cannot be parsed
-    """
-    invoice = invoice.lower().strip()
-
-    if not invoice.startswith("ln"):
-        raise LNURLError("Invalid Lightning invoice format")
-
-    # Find the network part (bc, tb, etc.)
-    network_start = 2
-    while network_start < len(invoice) and invoice[network_start] not in "0123456789":
-        network_start += 1
-
-    if network_start >= len(invoice):
-        raise LNURLError("Invalid Lightning invoice format")
-
-    # Parse amount and multiplier
-    amount_str = ""
-    multiplier = ""
-    i = network_start
-
-    # Extract numeric part
-    while i < len(invoice) and invoice[i].isdigit():
-        amount_str += invoice[i]
-        i += 1
-
-    # Extract multiplier if present
-    if i < len(invoice) and invoice[i] in "munp":
-        multiplier = invoice[i]
-        i += 1
-
-    # Check if we have the required "1" separator
-    if i >= len(invoice) or invoice[i] != "1":
-        raise LNURLError("Invalid Lightning invoice format")
-
-    if not amount_str:
-        raise LNURLError("Lightning invoice amount not specified")
-
-    # Convert to base units
-    try:
-        amount = int(amount_str)
-    except ValueError:
-        raise LNURLError("Invalid Lightning invoice amount")
-
-    # Apply multiplier to get millisatoshis
-    if multiplier == "m":  # milli = 10^-3
-        amount_msat = amount * 100_000_000  # amount is in BTC * 10^-3
-    elif multiplier == "u":  # micro = 10^-6
-        amount_msat = amount * 100_000  # amount is in BTC * 10^-6
-    elif multiplier == "n":  # nano = 10^-9
-        amount_msat = amount * 100  # amount is in BTC * 10^-9
-    elif multiplier == "p":  # pico = 10^-12
-        amount_msat = amount // 10  # amount is in BTC * 10^-12
-    else:
-        # No multiplier means the amount is in BTC
-        amount_msat = amount * 100_000_000_000  # Convert BTC to msat
-
-    # Convert to target currency unit
-    if currency == "msat":
-        return amount_msat
-    elif currency == "sat":
-        return amount_msat // 1000
-    else:
-        raise LNURLError(f"Unsupported currency for Lightning: {currency}")
-
-
 async def decode_lnurl(lnurl: str) -> str:
     """Decode LNURL to get the actual URL.
 
@@ -283,7 +207,7 @@ async def raw_send_to_lnurl(
             f"({min_sendable_sat} - {max_sendable_sat} {unit})"
         )
 
-    estimated_fees_sat = int(max(math.ceil((amount_msat / 1000) * 0.01), 2))
+    estimated_fees_sat = int(max(math.ceil((amount_msat / 1000) * 0.01), 2)) + 1
     estimated_fees_msat = estimated_fees_sat * 1000
     final_amount = amount_msat - estimated_fees_msat
 
@@ -291,9 +215,7 @@ async def raw_send_to_lnurl(
         lnurl_data["callback_url"], final_amount
     )
 
-    melt_quote_resp = await wallet.melt_quote(
-        invoice=bolt11_invoice, amount_msat=final_amount
-    )
+    melt_quote_resp = await wallet.melt_quote(invoice=bolt11_invoice)
 
     if amount:
         proofs, _ = await wallet.select_to_send(proofs, amount, set_reserved=True)

routstr/payment/models.py

@@ -143,14 +143,6 @@ async def async_fetch_openrouter_models(source_filter: str | None = None) -> lis
         return []
 
 
-def is_openrouter_upstream() -> bool:
-    try:
-        base = (settings.upstream_base_url or "").strip().rstrip("/")
-    except Exception:
-        return False
-    return base.lower() == "https://openrouter.ai/api/v1"
-
-
 def _row_to_model(
     row: ModelRow, apply_provider_fee: bool = False, provider_fee: float = 1.01
 ) -> Model:
@@ -203,33 +195,11 @@ def _row_to_model(
     return model
 
 
-def _model_to_row_payload(model: Model) -> dict[str, str | int | bool | None]:
-    return {
-        "id": model.id,
-        "name": model.name,
-        "created": model.created,
-        "description": model.description,
-        "context_length": model.context_length,
-        "architecture": json.dumps(model.architecture.dict()),
-        "pricing": json.dumps(model.pricing.dict()),
-        "sats_pricing": json.dumps(model.sats_pricing.dict())
-        if model.sats_pricing
-        else None,
-        "per_request_limits": json.dumps(model.per_request_limits)
-        if model.per_request_limits is not None
-        else None,
-        "top_provider": json.dumps(model.top_provider.dict())
-        if model.top_provider is not None
-        else None,
-        "enabled": model.enabled,
-        "upstream_provider_id": model.upstream_provider_id,
-    }
-
-
 async def list_models(
     session: AsyncSession,
     upstream_id: int,
     include_disabled: bool = False,
+    apply_fees: bool = True,
 ) -> list[Model]:
     from sqlmodel import select
 
@@ -247,28 +217,20 @@ async def list_models(
     return [
         _row_to_model(
             r,
-            apply_provider_fee=True,
+            apply_provider_fee=apply_fees,
             provider_fee=providers_by_id[r.upstream_provider_id].provider_fee
             if r.upstream_provider_id in providers_by_id
             else 1.01,
         )
         for r in rows
+        if include_disabled
+        or (
+            r.upstream_provider_id in providers_by_id
+            and providers_by_id[r.upstream_provider_id].enabled
+        )
     ]
 
 
-async def get_model_by_id(
-    model_id: str, provider_id: int, session: AsyncSession
-) -> Model | None:
-    from ..core.db import UpstreamProviderRow
-
-    row = await session.get(ModelRow, (model_id, provider_id))
-    if not row or not row.enabled:
-        return None
-    provider = await session.get(UpstreamProviderRow, provider_id)
-    provider_fee = provider.provider_fee if provider else 1.01
-    return _row_to_model(row, apply_provider_fee=True, provider_fee=provider_fee)
-
-
 def _calculate_usd_max_costs(model: Model) -> tuple[float, float, float]:
     """Calculate max costs in USD based on model context/token limits.
 

routstr/payment/price.py

@@ -79,15 +79,29 @@ async def _fetch_btc_usd_price() -> float:
     """Fetch the lowest BTC/USD price from multiple exchanges."""
     async with httpx.AsyncClient(timeout=30.0) as client:
         try:
-            prices = await asyncio.gather(
-                _kraken_btc_usd(client),
-                _coinbase_btc_usd(client),
-                _binance_btc_usdt(client),
-            )
-            valid_prices = [price for price in prices if price is not None]
+            tasks = [
+                asyncio.create_task(_kraken_btc_usd(client)),
+                asyncio.create_task(_coinbase_btc_usd(client)),
+                asyncio.create_task(_binance_btc_usdt(client)),
+            ]
+            valid_prices: list[float] = []
+
+            for future in asyncio.as_completed(tasks):
+                price = await future
+                if price is not None:
+                    valid_prices.append(price)
+
+                if len(valid_prices) >= 2:
+                    break
+
+            for task in tasks:
+                if not task.done():
+                    task.cancel()
+
             if not valid_prices:
                 logger.error("No valid BTC prices obtained from any exchange")
                 raise ValueError("Unable to fetch BTC price from any exchange")
+
             return min(valid_prices)
         except Exception as e:
             logger.error(

routstr/proxy.py

@@ -16,6 +16,7 @@ from .core.db import (
     create_session,
     get_session,
 )
+from .core.exceptions import UpstreamError
 from .core.settings import settings
 from .payment.helpers import (
     calculate_discounted_max_cost,
@@ -26,14 +27,15 @@ from .payment.helpers import (
 from .payment.models import Model
 from .upstream import BaseUpstreamProvider
 from .upstream.helpers import init_upstreams
-from .wallet import deserialize_token_from_string
 
 logger = get_logger(__name__)
 proxy_router = APIRouter()
 
 _upstreams: list[BaseUpstreamProvider] = []
 _model_instances: dict[str, Model] = {}  # All aliases -> Model
-_provider_map: dict[str, BaseUpstreamProvider] = {}  # All aliases -> Provider
+_provider_map: dict[
+    str, list[BaseUpstreamProvider]
+] = {}  # All aliases -> List[Provider]
 _unique_models: dict[str, Model] = {}  # Unique model.id -> Model (no duplicates)
 
 
@@ -70,8 +72,8 @@ def get_model_instance(model_id: str) -> Model | None:
     return _model_instances.get(model_id.lower())
 
 
-def get_provider_for_model(model_id: str) -> BaseUpstreamProvider | None:
-    """Get UpstreamProvider for model ID from global cache."""
+def get_provider_for_model(model_id: str) -> list[BaseUpstreamProvider] | None:
+    """Get UpstreamProvider list for model ID from global cache."""
     return _provider_map.get(model_id.lower())
 
 
@@ -98,6 +100,8 @@ async def refresh_model_maps() -> None:
     disabled_model_ids: set[str] = set()
 
     for provider in provider_rows:
+        if not provider.enabled:
+            continue
         for model in provider.models:
             if model.enabled:
                 overrides_by_id[model.id] = (model, provider.provider_fee)
@@ -148,32 +152,14 @@ async def proxy(
     else:
         model_id = request_body_dict.get("model", "unknown")
 
-    if "https://testnut.cashu.space" in settings.cashu_mints:
-        try:
-            token_str = None
-            if x_cashu_header := headers.get("x-cashu"):
-                token_str = x_cashu_header
-            elif auth_header := headers.get("authorization"):
-                parts = auth_header.split(" ")
-                if len(parts) > 1 and not parts[1].startswith("sk-"):
-                    token_str = parts[1]
-
-            if token_str:
-                token_obj = deserialize_token_from_string(token_str)
-                if token_obj.mint == "https://testnut.cashu.space":
-                    model_id = "mock/gpt-420-mock"
-                    request_body_dict["model"] = model_id
-        except Exception:
-            pass
-
     model_obj = get_model_instance(model_id)
     if not model_obj:
         return create_error_response(
             "invalid_model", f"Model '{model_id}' not found", 400, request=request
         )
 
-    upstream = get_provider_for_model(model_id)
-    if not upstream:
+    upstreams = get_provider_for_model(model_id)
+    if not upstreams:
         return create_error_response(
             "invalid_model",
             f"No provider found for model '{model_id}'",
@@ -181,26 +167,52 @@ async def proxy(
             request=request,
         )
 
+    # todo figure out cost calculation since fallback provider is usually not the same price
+    # Use first provider for initial checks/cost calculation
+    # primary_upstream = upstreams[0]
+
     _max_cost_for_model = await get_max_cost_for_model(
         model=model_id, session=session, model_obj=model_obj
     )
     max_cost_for_model = await calculate_discounted_max_cost(
         _max_cost_for_model, request_body_dict, model_obj=model_obj
     )
+    # Ensure max_cost_for_model is at least the minimum allowed request cost
+    max_cost_for_model = max(max_cost_for_model, settings.min_request_msat)
+
     check_token_balance(headers, request_body_dict, max_cost_for_model)
 
     if x_cashu := headers.get("x-cashu", None):
-        if is_responses_api:
-            return await upstream.handle_x_cashu_responses(
-                request, x_cashu, path, max_cost_for_model, model_obj
-            )
-        else:
-            return await upstream.handle_x_cashu(
-                request, x_cashu, path, max_cost_for_model, model_obj
-            )
+        last_error = None
+        for i, upstream in enumerate(upstreams):
+            try:
+                if is_responses_api:
+                    return await upstream.handle_x_cashu_responses(
+                        request, x_cashu, path, max_cost_for_model, model_obj
+                    )
+                else:
+                    return await upstream.handle_x_cashu(
+                        request, x_cashu, path, max_cost_for_model, model_obj
+                    )
+            except UpstreamError as e:
+                logger.warning(
+                    f"Upstream {upstream.provider_type} failed (x-cashu): {e}"
+                )
+                if i == len(upstreams) - 1:
+                    last_error = e
+                continue
+
+        return create_error_response(
+            "upstream_error",
+            str(last_error) if last_error else "All upstreams failed",
+            502,
+            request=request,
+        )
 
     elif auth := headers.get("authorization", None):
-        key = await get_bearer_token_key(headers, path, session, auth)
+        key = await get_bearer_token_key(
+            headers, path, session, auth, max_cost_for_model
+        )
 
     else:
         if request.method not in ["GET"]:
@@ -212,60 +224,170 @@ async def proxy(
             )
 
         logger.debug("Processing unauthenticated GET request", extra={"path": path})
-        headers = upstream.prepare_headers(dict(request.headers))
-        return await upstream.forward_get_request(request, path, headers)
+
+        last_error_response = None
+        for i, upstream in enumerate(upstreams):
+            try:
+                headers = upstream.prepare_headers(dict(request.headers))
+                response = await upstream.forward_get_request(request, path, headers)
+
+                if response.status_code in [502, 429] and i < len(upstreams) - 1:
+                    error_message = ""
+                    try:
+                        if hasattr(response, "body"):
+                            body_bytes = response.body
+                            data = json.loads(body_bytes)
+                            if "error" in data:
+                                error_data = data["error"]
+                                if isinstance(error_data, dict):
+                                    error_message = error_data.get("message", "")
+                                elif isinstance(error_data, str):
+                                    error_message = error_data
+                    except Exception:
+                        pass
+
+                    await upstream.on_upstream_error_redirect(
+                        response.status_code, error_message
+                    )
+
+                    logger.warning(
+                        f"Upstream {upstream.provider_type} returned {response.status_code} (GET), trying next provider",
+                        extra={
+                            "status_code": response.status_code,
+                            "upstream": upstream.provider_type,
+                        },
+                    )
+                    continue
+                return response
+            except UpstreamError as e:
+                logger.warning(f"Upstream {upstream.provider_type} failed (GET): {e}")
+                if i == len(upstreams) - 1:
+                    last_error_response = create_error_response(
+                        "upstream_error", str(e), 502, request=request
+                    )
+                continue
+        return last_error_response or create_error_response(
+            "upstream_error", "All upstreams failed", 502, request=request
+        )
 
     if request_body_dict:
         await pay_for_request(key, max_cost_for_model, session)
 
-    headers = upstream.prepare_headers(dict(request.headers))
+    for i, upstream in enumerate(upstreams):
+        headers = upstream.prepare_headers(dict(request.headers))
 
-    if is_responses_api:
-        response = await upstream.forward_responses_request(
-            request,
-            path,
-            headers,
-            request_body,
-            key,
-            max_cost_for_model,
-            session,
-            model_obj,
-        )
-    else:
-        response = await upstream.forward_request(
-            request,
-            path,
-            headers,
-            request_body,
-            key,
-            max_cost_for_model,
-            session,
-            model_obj,
-        )
+        try:
+            try:
+                if is_responses_api:
+                    response = await upstream.forward_responses_request(
+                        request,
+                        path,
+                        headers,
+                        request_body,
+                        key,
+                        max_cost_for_model,
+                        session,
+                        model_obj,
+                    )
+                else:
+                    response = await upstream.forward_request(
+                        request,
+                        path,
+                        headers,
+                        request_body,
+                        key,
+                        max_cost_for_model,
+                        session,
+                        model_obj,
+                    )
+            except Exception as e:
+                logger.error(
+                    "Upstream request failed, ensuring payment is reverted",
+                    extra={
+                        "error": str(e),
+                        "error_type": type(e).__name__,
+                        "path": path,
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "max_cost_for_model": max_cost_for_model,
+                    },
+                )
+                await revert_pay_for_request(key, session, max_cost_for_model)
+                raise
 
-    if response.status_code != 200:
-        await revert_pay_for_request(key, session, max_cost_for_model)
-        logger.warning(
-            "Upstream request failed, revert payment",
-            extra={
-                "status_code": response.status_code,
-                "path": path,
-                "key_hash": key.hashed_key[:8] + "...",
-                "key_balance": key.balance,
-                "max_cost_for_model": max_cost_for_model,
-                "upstream_headers": response.headers
-                if hasattr(response, "headers")
-                else None,
-            },
-        )
-        # Return the mapped error response generated earlier rather than masking with 502
-        return response
+            if response.status_code != 200:
+                # Check if we should retry (502 Upstream Error or 429 Rate Limit)
+                should_retry = response.status_code in [502, 429, 400, 401, 403, 404]
+                if should_retry and i < len(upstreams) - 1:
+                    error_message = ""
+                    try:
+                        if hasattr(response, "body"):
+                            body_bytes = response.body
+                            data = json.loads(body_bytes)
+                            if "error" in data:
+                                error_data = data["error"]
+                                if isinstance(error_data, dict):
+                                    error_message = error_data.get("message", "")
+                                elif isinstance(error_data, str):
+                                    error_message = error_data
+                    except Exception:
+                        pass
 
-    return response
+                    await upstream.on_upstream_error_redirect(
+                        response.status_code, error_message
+                    )
+
+                    logger.warning(
+                        f"Upstream {upstream.provider_type} returned {response.status_code}, trying next provider",
+                        extra={
+                            "status_code": response.status_code,
+                            "upstream": upstream.provider_type,
+                        },
+                    )
+                    continue
+
+                # 4xx error (user error), or other non-retryable error, or last provider failed
+                await revert_pay_for_request(key, session, max_cost_for_model)
+                logger.warning(
+                    "Upstream request failed, revert payment",
+                    extra={
+                        "status_code": response.status_code,
+                        "path": path,
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "key_balance": key.balance,
+                        "max_cost_for_model": max_cost_for_model,
+                        "upstream_headers": response.headers
+                        if hasattr(response, "headers")
+                        else None,
+                    },
+                )
+                return response
+
+            return response
+
+        except UpstreamError as e:
+            logger.warning(
+                f"Upstream {upstream.provider_type} failed: {e}",
+                extra={"retry": i < len(upstreams) - 1},
+            )
+
+            # If this was the last provider
+            if i == len(upstreams) - 1:
+                await revert_pay_for_request(key, session, max_cost_for_model)
+                return create_error_response(
+                    "upstream_error", str(e), 502, request=request
+                )
+
+            # Otherwise loop continues to next provider
+            continue
+
+    # Should not be reached given logic above
+    return create_error_response(
+        "upstream_error", "All upstreams failed", 502, request=request
+    )
 
 
 async def get_bearer_token_key(
-    headers: dict, path: str, session: AsyncSession, auth: str
+    headers: dict, path: str, session: AsyncSession, auth: str, min_cost: int = 0
 ) -> ApiKey:
     """Handle bearer token authentication proxy requests."""
     bearer_key = auth.replace("Bearer ", "") if auth.startswith("Bearer ") else ""
@@ -281,6 +403,7 @@ async def get_bearer_token_key(
             "bearer_key_preview": bearer_key[:20] + "..."
             if len(bearer_key) > 20
             else bearer_key,
+            "min_cost": min_cost,
         },
     )
 
@@ -319,6 +442,7 @@ async def get_bearer_token_key(
             session,
             refund_address,
             key_expiry_time,  # type: ignore
+            min_cost=min_cost,
         )
         logger.info(
             "Bearer token validated successfully",
@@ -357,7 +481,7 @@ def extract_model_from_responses_request(request_body_dict: dict[str, Any]) -> s
 
     logger.warning(
         "No model found in Responses API request",
-        extra={"body_keys": list(request_body_dict.keys())}
+        extra={"body_keys": list(request_body_dict.keys())},
     )
     return "unknown"
 

routstr/upstream/base.py

@@ -12,9 +12,10 @@ from fastapi import BackgroundTasks, HTTPException, Request
 from fastapi.responses import Response, StreamingResponse
 from pydantic import BaseModel
 
-from ..auth import adjust_payment_for_tokens
+from ..auth import adjust_payment_for_tokens, revert_pay_for_request
 from ..core import get_logger
 from ..core.db import ApiKey, AsyncSession, create_session
+from ..core.exceptions import UpstreamError
 
 if TYPE_CHECKING:
     from ..core.db import UpstreamProviderRow
@@ -234,7 +235,11 @@ class BaseUpstreamProvider:
                     )
 
                 # Handle model in input field (alternative format)
-                if "input" in data and isinstance(data["input"], dict) and "model" in data["input"]:
+                if (
+                    "input" in data
+                    and isinstance(data["input"], dict)
+                    and "model" in data["input"]
+                ):
                     original_model = model_obj.id
                     transformed_model = self.transform_model_name(original_model)
                     data["input"]["model"] = transformed_model
@@ -336,6 +341,20 @@ class BaseUpstreamProvider:
                 message = preview[:500]
         return message, upstream_code
 
+    async def on_upstream_error_redirect(
+        self, status_code: int, error_message: str
+    ) -> None:
+        """Hook called when the proxy redirects to another provider due to an error.
+
+        Subclasses can implement this to perform actions like disabling the provider
+        if it's out of balance.
+
+        Args:
+            status_code: The HTTP status code returned by the upstream
+            error_message: The error message extracted from the upstream response
+        """
+        pass
+
     async def map_upstream_error_response(
         self, request: Request, path: str, upstream_response: httpx.Response
     ) -> Response:
@@ -408,7 +427,11 @@ class BaseUpstreamProvider:
         )
 
     async def handle_streaming_chat_completion(
-        self, response: httpx.Response, key: ApiKey, max_cost_for_model: int
+        self,
+        response: httpx.Response,
+        key: ApiKey,
+        max_cost_for_model: int,
+        background_tasks: BackgroundTasks,
     ) -> StreamingResponse:
         """Handle streaming chat completion responses with token usage tracking and cost adjustment.
 
@@ -432,156 +455,120 @@ class BaseUpstreamProvider:
         async def stream_with_cost(
             max_cost_for_model: int,
         ) -> AsyncGenerator[bytes, None]:
-            stored_chunks: list[bytes] = []
             usage_finalized: bool = False
             last_model_seen: str | None = None
+            usage_chunk_data: dict | None = None
+            done_seen: bool = False
 
-            async def finalize_without_usage() -> bytes | None:
+            async def finalize_db_only() -> None:
                 nonlocal usage_finalized
                 if usage_finalized:
-                    return None
+                    return
                 async with create_session() as new_session:
                     fresh_key = await new_session.get(key.__class__, key.hashed_key)
                     if not fresh_key:
-                        return None
+                        return
                     try:
-                        fallback: dict = {
-                            "model": last_model_seen or "unknown",
-                            "usage": None,
-                        }
-                        cost_data = await adjust_payment_for_tokens(
-                            fresh_key, fallback, new_session, max_cost_for_model
+                        await adjust_payment_for_tokens(
+                            fresh_key,
+                            {"model": last_model_seen or "unknown", "usage": None},
+                            new_session,
+                            max_cost_for_model,
                         )
                         usage_finalized = True
-                        logger.info(
-                            "Finalized streaming payment without explicit usage",
-                            extra={
-                                "key_hash": key.hashed_key[:8] + "...",
-                                "cost_data": cost_data,
-                                "balance_after_adjustment": fresh_key.balance,
-                            },
-                        )
-                        return f"data: {json.dumps({'cost': cost_data})}\n\n".encode()
-                    except Exception as cost_error:
-                        logger.error(
-                            "Error finalizing payment without usage",
-                            extra={
-                                "error": str(cost_error),
-                                "error_type": type(cost_error).__name__,
-                                "key_hash": key.hashed_key[:8] + "...",
-                            },
-                        )
-                        return None
-
-            try:
-                async for chunk in response.aiter_bytes():
-                    stored_chunks.append(chunk)
-                    try:
-                        for part in re.split(b"data: ", chunk):
-                            if not part or part.strip() in (b"[DONE]", b""):
-                                continue
-                            try:
-                                obj = json.loads(part)
-                                if isinstance(obj, dict) and obj.get("model"):
-                                    last_model_seen = str(obj.get("model"))
-                            except json.JSONDecodeError:
-                                pass
                     except Exception:
                         pass
 
-                    yield chunk
+            try:
+                async for chunk in response.aiter_bytes():
+                    # Split chunk into SSE events
+                    parts = re.split(b"data: ", chunk)
+                    for i, part in enumerate(parts):
+                        if not part:
+                            continue
 
-                logger.debug(
-                    "Streaming completed, analyzing usage data",
-                    extra={
-                        "key_hash": key.hashed_key[:8] + "...",
-                        "chunks_count": len(stored_chunks),
-                    },
-                )
+                        stripped_part = part.strip()
+                        if not stripped_part:
+                            continue
 
-                for i in range(len(stored_chunks) - 1, -1, -1):
-                    chunk = stored_chunks[i]
-                    if not chunk:
-                        continue
-                    try:
-                        events = re.split(b"data: ", chunk)
-                        for event_data in events:
-                            if not event_data or event_data.strip() in (b"[DONE]", b""):
-                                continue
-                            try:
-                                data = json.loads(event_data)
-                                if isinstance(data, dict) and data.get("model"):
-                                    last_model_seen = str(data.get("model"))
-                                if isinstance(data, dict) and isinstance(
-                                    data.get("usage"), dict
-                                ):
-                                    async with create_session() as new_session:
-                                        fresh_key = await new_session.get(
-                                            key.__class__, key.hashed_key
-                                        )
-                                        if fresh_key:
-                                            try:
-                                                cost_data = (
-                                                    await adjust_payment_for_tokens(
-                                                        fresh_key,
-                                                        data,
-                                                        new_session,
-                                                        max_cost_for_model,
-                                                    )
-                                                )
-                                                usage_finalized = True
-                                                logger.info(
-                                                    "Payment adjustment completed for streaming",
-                                                    extra={
-                                                        "key_hash": key.hashed_key[:8]
-                                                        + "...",
-                                                        "cost_data": cost_data,
-                                                        "model": last_model_seen,
-                                                        "balance_after_adjustment": fresh_key.balance,
-                                                    },
-                                                )
-                                                yield f"data: {json.dumps({'cost': cost_data})}\n\n".encode()
-                                            except Exception as cost_error:
-                                                logger.error(
-                                                    "Error adjusting payment for streaming tokens",
-                                                    extra={
-                                                        "error": str(cost_error),
-                                                        "error_type": type(
-                                                            cost_error
-                                                        ).__name__,
-                                                        "key_hash": key.hashed_key[:8]
-                                                        + "...",
-                                                    },
-                                                )
-                                    break
-                            except json.JSONDecodeError:
-                                continue
-                    except Exception as e:
-                        logger.error(
-                            "Error processing streaming response chunk",
-                            extra={
-                                "error": str(e),
-                                "error_type": type(e).__name__,
-                                "key_hash": key.hashed_key[:8] + "...",
-                            },
+                        if stripped_part == b"[DONE]":
+                            done_seen = True
+                            continue
+
+                        try:
+                            obj = json.loads(part)
+                            if isinstance(obj, dict):
+                                if obj.get("model"):
+                                    last_model_seen = str(obj.get("model"))
+
+                                if isinstance(obj.get("usage"), dict):
+                                    # Hold this chunk back to merge cost later
+                                    usage_chunk_data = obj
+                                    continue
+                        except json.JSONDecodeError:
+                            pass
+
+                        prefix = (
+                            b"data: " if (i > 0 or chunk.startswith(b"data: ")) else b""
                         )
+                        yield prefix + part
+
+                # Stream finished, process usage if found
+                if usage_chunk_data:
+                    async with create_session() as session:
+                        fresh_key = await session.get(key.__class__, key.hashed_key)
+                        if fresh_key:
+                            try:
+                                cost_data = await adjust_payment_for_tokens(
+                                    fresh_key,
+                                    usage_chunk_data,
+                                    session,
+                                    max_cost_for_model,
+                                )
+                                # Merge cost into usage
+                                usage_chunk_data["usage"]["cost"] = cost_data.get(
+                                    "total_usd", 0.0
+                                )
+                                # Keep detailed cost in metadata
+                                usage_chunk_data["metadata"] = usage_chunk_data.get(
+                                    "metadata", {}
+                                )
+                                usage_chunk_data["metadata"]["routstr"] = {
+                                    "cost": cost_data
+                                }
+                                yield f"data: {json.dumps(usage_chunk_data)}\n\n".encode()
+                                usage_finalized = True
+                            except Exception as e:
+                                logger.exception(
+                                    "Error during usage finalization",
+                                    extra={
+                                        "key_hash": key.hashed_key[:8] + "...",
+                                        "error": str(e),
+                                    },
+                                )
+                                # Fallback: yield original usage chunk if adjustment fails
+                                yield f"data: {json.dumps(usage_chunk_data)}\n\n".encode()
 
                 if not usage_finalized:
-                    maybe_cost_event = await finalize_without_usage()
-                    if maybe_cost_event is not None:
-                        yield maybe_cost_event
+                    await finalize_db_only()
+
+                if done_seen:
+                    yield b"data: [DONE]\n\n"
 
             except Exception as stream_error:
                 logger.warning(
-                    "Streaming interrupted; finalizing without usage",
+                    "Streaming interrupted; finalizing in background",
                     extra={
                         "error": str(stream_error),
-                        "error_type": type(stream_error).__name__,
                         "key_hash": key.hashed_key[:8] + "...",
                     },
                 )
-                await finalize_without_usage()
                 raise
+            finally:
+                if not usage_finalized:
+                    # Create a background task to ensure finalization happens
+                    # even if the generator is closed early
+                    background_tasks.add_task(finalize_db_only)
 
         # Remove inaccurate encoding headers from upstream response
         response_headers = dict(response.headers)
@@ -621,6 +608,7 @@ class BaseUpstreamProvider:
             },
         )
 
+        content: bytes | None = None
         try:
             content = await response.aread()
             response_json = json.loads(content)
@@ -637,6 +625,14 @@ class BaseUpstreamProvider:
             cost_data = await adjust_payment_for_tokens(
                 key, response_json, session, deducted_max_cost
             )
+
+            # Merge cost into usage for OpenCode
+            if "usage" in response_json:
+                response_json["usage"]["cost"] = cost_data.get("total_usd", 0.0)
+
+            # Keep detailed cost
+            response_json["metadata"] = response_json.get("metadata", {})
+            response_json["metadata"]["routstr"] = {"cost": cost_data}
             response_json["cost"] = cost_data
 
             logger.info(
@@ -722,167 +718,135 @@ class BaseUpstreamProvider:
         async def stream_with_responses_cost(
             max_cost_for_model: int,
         ) -> AsyncGenerator[bytes, None]:
-            stored_chunks: list[bytes] = []
             usage_finalized: bool = False
             last_model_seen: str | None = None
             reasoning_tokens: int = 0
+            usage_chunk_data: dict | None = None
+            done_seen: bool = False
 
-            async def finalize_without_usage() -> bytes | None:
+            async def finalize_db_only() -> None:
                 nonlocal usage_finalized
                 if usage_finalized:
-                    return None
+                    return
                 async with create_session() as new_session:
                     fresh_key = await new_session.get(key.__class__, key.hashed_key)
                     if not fresh_key:
-                        return None
+                        return
                     try:
-                        fallback: dict = {
-                            "model": last_model_seen or "unknown",
-                            "usage": None,
-                        }
-                        cost_data = await adjust_payment_for_tokens(
-                            fresh_key, fallback, new_session, max_cost_for_model
+                        await adjust_payment_for_tokens(
+                            fresh_key,
+                            {"model": last_model_seen or "unknown", "usage": None},
+                            new_session,
+                            max_cost_for_model,
                         )
                         usage_finalized = True
-                        logger.info(
-                            "Finalized Responses API streaming payment without explicit usage",
-                            extra={
-                                "key_hash": key.hashed_key[:8] + "...",
-                                "cost_data": cost_data,
-                                "balance_after_adjustment": fresh_key.balance,
-                            },
-                        )
-                        return f"data: {json.dumps({'cost': cost_data})}\\n\\n".encode()
-                    except Exception as cost_error:
-                        logger.error(
-                            "Error finalizing Responses API payment without usage",
-                            extra={
-                                "error": str(cost_error),
-                                "error_type": type(cost_error).__name__,
-                                "key_hash": key.hashed_key[:8] + "...",
-                            },
-                        )
-                        return None
-
-            try:
-                async for chunk in response.aiter_bytes():
-                    stored_chunks.append(chunk)
-                    try:
-                        for part in re.split(b"data: ", chunk):
-                            if not part or part.strip() in (b"[DONE]", b""):
-                                continue
-                            try:
-                                obj = json.loads(part)
-                                if isinstance(obj, dict):
-                                    if obj.get("model"):
-                                        last_model_seen = str(obj.get("model"))
-
-                                    # Track reasoning tokens for Responses API
-                                    if usage := obj.get("usage", {}):
-                                        if isinstance(usage, dict) and "reasoning_tokens" in usage:
-                                            reasoning_tokens += usage.get("reasoning_tokens", 0)
-                            except json.JSONDecodeError:
-                                pass
                     except Exception:
                         pass
 
-                    yield chunk
+            try:
+                async for chunk in response.aiter_bytes():
+                    # Split chunk into SSE events
+                    parts = re.split(b"data: ", chunk)
+                    for i, part in enumerate(parts):
+                        if not part:
+                            continue
 
-                logger.debug(
-                    "Responses API streaming completed, analyzing usage data",
-                    extra={
-                        "key_hash": key.hashed_key[:8] + "...",
-                        "chunks_count": len(stored_chunks),
-                        "reasoning_tokens": reasoning_tokens,
-                    },
-                )
+                        stripped_part = part.strip()
+                        if not stripped_part:
+                            continue
 
-                # Process final usage data
-                for i in range(len(stored_chunks) - 1, -1, -1):
-                    chunk = stored_chunks[i]
-                    if not chunk:
-                        continue
-                    try:
-                        events = re.split(b"data: ", chunk)
-                        for event_data in events:
-                            if not event_data or event_data.strip() in (b"[DONE]", b""):
-                                continue
-                            try:
-                                data = json.loads(event_data)
-                                if isinstance(data, dict) and data.get("model"):
-                                    last_model_seen = str(data.get("model"))
-                                if isinstance(data, dict) and isinstance(
-                                    data.get("usage"), dict
-                                ):
-                                    # Include reasoning tokens in usage calculation
-                                    async with create_session() as new_session:
-                                        fresh_key = await new_session.get(
-                                            key.__class__, key.hashed_key
+                        if stripped_part == b"[DONE]":
+                            done_seen = True
+                            continue
+
+                        try:
+                            obj = json.loads(part)
+                            if isinstance(obj, dict):
+                                if obj.get("model"):
+                                    last_model_seen = str(obj.get("model"))
+
+                                # Track reasoning tokens for Responses API
+                                if usage := obj.get("usage", {}):
+                                    if (
+                                        isinstance(usage, dict)
+                                        and "reasoning_tokens" in usage
+                                    ):
+                                        reasoning_tokens += usage.get(
+                                            "reasoning_tokens", 0
                                         )
-                                        if fresh_key:
-                                            try:
-                                                cost_data = (
-                                                    await adjust_payment_for_tokens(
-                                                        fresh_key,
-                                                        data,
-                                                        new_session,
-                                                        max_cost_for_model,
-                                                    )
-                                                )
-                                                usage_finalized = True
-                                                logger.info(
-                                                    "Payment adjustment completed for Responses API streaming",
-                                                    extra={
-                                                        "key_hash": key.hashed_key[:8]
-                                                        + "...",
-                                                        "cost_data": cost_data,
-                                                        "model": last_model_seen,
-                                                        "reasoning_tokens": reasoning_tokens,
-                                                        "balance_after_adjustment": fresh_key.balance,
-                                                    },
-                                                )
-                                                yield f"data: {json.dumps({'cost': cost_data})}\\n\\n".encode()
-                                            except Exception as cost_error:
-                                                logger.error(
-                                                    "Error adjusting payment for Responses API streaming tokens",
-                                                    extra={
-                                                        "error": str(cost_error),
-                                                        "error_type": type(
-                                                            cost_error
-                                                        ).__name__,
-                                                        "key_hash": key.hashed_key[:8]
-                                                        + "...",
-                                                    },
-                                                )
-                                    break
-                            except json.JSONDecodeError:
-                                continue
-                    except Exception as e:
-                        logger.error(
-                            "Error processing Responses API streaming response chunk",
-                            extra={
-                                "error": str(e),
-                                "error_type": type(e).__name__,
-                                "key_hash": key.hashed_key[:8] + "...",
-                            },
+
+                                # Responses API usage is in response.completed/incomplete events
+                                chunk_type = obj.get("type", "")
+                                if chunk_type in (
+                                    "response.completed",
+                                    "response.incomplete",
+                                ):
+                                    usage_chunk_data = obj
+                                    continue
+                        except json.JSONDecodeError:
+                            pass
+
+                        prefix = (
+                            b"data: " if (i > 0 or chunk.startswith(b"data: ")) else b""
                         )
+                        yield prefix + part
+
+                # Stream finished, process usage if found
+                if usage_chunk_data:
+                    async with create_session() as session:
+                        fresh_key = await session.get(key.__class__, key.hashed_key)
+                        if fresh_key:
+                            try:
+                                cost_data = await adjust_payment_for_tokens(
+                                    fresh_key,
+                                    usage_chunk_data,
+                                    session,
+                                    max_cost_for_model,
+                                )
+                                # Merge cost into usage chunk
+                                if (
+                                    "response" in usage_chunk_data
+                                    and "usage" in usage_chunk_data["response"]
+                                ):
+                                    usage_chunk_data["response"]["usage"]["cost"] = (
+                                        cost_data.get("total_usd", 0.0)
+                                    )
+                                elif "usage" in usage_chunk_data:
+                                    usage_chunk_data["usage"]["cost"] = cost_data.get(
+                                        "total_usd", 0.0
+                                    )
+
+                                # Keep detailed cost in metadata
+                                usage_chunk_data["metadata"] = usage_chunk_data.get(
+                                    "metadata", {}
+                                )
+                                usage_chunk_data["metadata"]["routstr"] = {
+                                    "cost": cost_data
+                                }
+                                yield f"data: {json.dumps(usage_chunk_data)}\n\n".encode()
+                                usage_finalized = True
+                            except Exception:
+                                # Fallback: yield original usage chunk if adjustment fails
+                                yield f"data: {json.dumps(usage_chunk_data)}\n\n".encode()
 
                 if not usage_finalized:
-                    maybe_cost_event = await finalize_without_usage()
-                    if maybe_cost_event is not None:
-                        yield maybe_cost_event
+                    await finalize_db_only()
+
+                if done_seen:
+                    yield b"data: [DONE]\n\n"
 
             except Exception as stream_error:
                 logger.warning(
-                    "Responses API streaming interrupted; finalizing without usage",
+                    "Responses API streaming interrupted; finalizing in background",
                     extra={
                         "error": str(stream_error),
-                        "error_type": type(stream_error).__name__,
                         "key_hash": key.hashed_key[:8] + "...",
                     },
                 )
-                await finalize_without_usage()
                 raise
+            finally:
+                if not usage_finalized:
+                    await finalize_db_only()
 
         # Remove inaccurate encoding headers from upstream response
         response_headers = dict(response.headers)
@@ -922,6 +886,7 @@ class BaseUpstreamProvider:
             },
         )
 
+        content: bytes | None = None
         try:
             content = await response.aread()
             response_json = json.loads(content)
@@ -933,14 +898,22 @@ class BaseUpstreamProvider:
                     "model": response_json.get("model", "unknown"),
                     "has_usage": "usage" in response_json,
                     "has_reasoning_tokens": "usage" in response_json
-                        and isinstance(response_json.get("usage"), dict)
-                        and "reasoning_tokens" in response_json["usage"],
+                    and isinstance(response_json.get("usage"), dict)
+                    and "reasoning_tokens" in response_json["usage"],
                 },
             )
 
             cost_data = await adjust_payment_for_tokens(
                 key, response_json, session, deducted_max_cost
             )
+
+            # Merge cost into usage for OpenCode
+            if "usage" in response_json:
+                response_json["usage"]["cost"] = cost_data.get("total_usd", 0.0)
+
+            # Keep detailed cost
+            response_json["metadata"] = response_json.get("metadata", {})
+            response_json["metadata"]["routstr"] = {"cost": cost_data}
             response_json["cost"] = cost_data
 
             logger.info(
@@ -1001,6 +974,44 @@ class BaseUpstreamProvider:
             )
             raise
 
+    async def _finalize_generic_streaming_payment(
+        self, key_hash: str, max_cost: int, path: str
+    ) -> None:
+        """Background task to finalize payment for generic streaming requests."""
+        async with create_session() as session:
+            key = await session.get(ApiKey, key_hash)
+            if not key:
+                logger.warning(
+                    "Key not found during background payment finalization",
+                    extra={"key_hash": key_hash[:8] + "..."},
+                )
+                return
+
+            try:
+                # Finalize with "unknown" model and no usage to release reservation/charge max cost
+                await adjust_payment_for_tokens(
+                    key,
+                    {"model": "unknown", "usage": None},
+                    session,
+                    max_cost,
+                )
+                logger.info(
+                    "Finalized generic streaming payment in background",
+                    extra={
+                        "path": path,
+                        "key_hash": key_hash[:8] + "...",
+                    },
+                )
+            except Exception as e:
+                logger.error(
+                    "Error finalizing generic streaming payment in background",
+                    extra={
+                        "error": str(e),
+                        "key_hash": key_hash[:8] + "...",
+                        "path": path,
+                    },
+                )
+
     async def forward_request(
         self,
         request: Request,
@@ -1085,6 +1096,14 @@ class BaseUpstreamProvider:
             )
 
             if response.status_code != 200:
+                if response.status_code >= 500:
+                    await response.aclose()
+                    await client.aclose()
+                    raise UpstreamError(
+                        f"Upstream returned status {response.status_code}",
+                        status_code=response.status_code,
+                    )
+
                 try:
                     mapped_error = await self.map_upstream_error_response(
                         request, path, response
@@ -1130,12 +1149,12 @@ class BaseUpstreamProvider:
                     )
 
                     if is_streaming and response.status_code == 200:
-                        result = await self.handle_streaming_chat_completion(
-                            response, key, max_cost_for_model
-                        )
                         background_tasks = BackgroundTasks()
                         background_tasks.add_task(response.aclose)
                         background_tasks.add_task(client.aclose)
+                        result = await self.handle_streaming_chat_completion(
+                            response, key, max_cost_for_model, background_tasks
+                        )
                         result.background = background_tasks
                         return result
 
@@ -1152,6 +1171,12 @@ class BaseUpstreamProvider:
             background_tasks = BackgroundTasks()
             background_tasks.add_task(response.aclose)
             background_tasks.add_task(client.aclose)
+            background_tasks.add_task(
+                self._finalize_generic_streaming_payment,
+                key.hashed_key,
+                max_cost_for_model,
+                path,
+            )
 
             logger.debug(
                 "Streaming non-chat response",
@@ -1169,6 +1194,9 @@ class BaseUpstreamProvider:
                 background=background_tasks,
             )
 
+        except UpstreamError:
+            raise
+
         except httpx.RequestError as exc:
             await client.aclose()
             error_type = type(exc).__name__
@@ -1187,6 +1215,8 @@ class BaseUpstreamProvider:
                 },
             )
 
+            await revert_pay_for_request(key, session, max_cost_for_model)
+
             if isinstance(exc, httpx.ConnectError):
                 error_message = "Unable to connect to upstream service"
             elif isinstance(exc, httpx.TimeoutException):
@@ -1196,9 +1226,7 @@ class BaseUpstreamProvider:
             else:
                 error_message = f"Error connecting to upstream service: {error_type}"
 
-            return create_error_response(
-                "upstream_error", error_message, 502, request=request
-            )
+            raise UpstreamError(error_message, status_code=502)
 
         except Exception as exc:
             await client.aclose()
@@ -1218,6 +1246,8 @@ class BaseUpstreamProvider:
                 },
             )
 
+            await revert_pay_for_request(key, session, max_cost_for_model)
+
             return create_error_response(
                 "internal_error",
                 "An unexpected server error occurred",
@@ -1311,6 +1341,14 @@ class BaseUpstreamProvider:
             )
 
             if response.status_code != 200:
+                if response.status_code >= 500:
+                    await response.aclose()
+                    await client.aclose()
+                    raise UpstreamError(
+                        f"Upstream returned status {response.status_code}",
+                        status_code=response.status_code,
+                    )
+
                 try:
                     mapped_error = await self.map_upstream_error_response(
                         request, path, response
@@ -1355,9 +1393,15 @@ class BaseUpstreamProvider:
             background_tasks = BackgroundTasks()
             background_tasks.add_task(response.aclose)
             background_tasks.add_task(client.aclose)
+            background_tasks.add_task(
+                self._finalize_generic_streaming_payment,
+                key.hashed_key,
+                max_cost_for_model,
+                path,
+            )
 
             logger.debug(
-                "Streaming non-chat response",
+                "Streaming non-Responses API response",
                 extra={
                     "path": path,
                     "status_code": response.status_code,
@@ -1372,6 +1416,9 @@ class BaseUpstreamProvider:
                 background=background_tasks,
             )
 
+        except UpstreamError:
+            raise
+
         except httpx.RequestError as exc:
             await client.aclose()
             error_type = type(exc).__name__
@@ -1390,6 +1437,8 @@ class BaseUpstreamProvider:
                 },
             )
 
+            await revert_pay_for_request(key, session, max_cost_for_model)
+
             if isinstance(exc, httpx.ConnectError):
                 error_message = "Unable to connect to upstream service"
             elif isinstance(exc, httpx.TimeoutException):
@@ -1399,9 +1448,7 @@ class BaseUpstreamProvider:
             else:
                 error_message = f"Error connecting to upstream service: {error_type}"
 
-            return create_error_response(
-                "upstream_error", error_message, 502, request=request
-            )
+            raise UpstreamError(error_message, status_code=502)
 
         except Exception as exc:
             await client.aclose()
@@ -1421,6 +1468,8 @@ class BaseUpstreamProvider:
                 },
             )
 
+            await revert_pay_for_request(key, session, max_cost_for_model)
+
             return create_error_response(
                 "internal_error",
                 "An unexpected server error occurred",
@@ -2503,7 +2552,10 @@ class BaseUpstreamProvider:
                         usage_data = data_json["usage"]
                         model = data_json.get("model")
                         # Track reasoning tokens for Responses API
-                        if isinstance(usage_data, dict) and "reasoning_tokens" in usage_data:
+                        if (
+                            isinstance(usage_data, dict)
+                            and "reasoning_tokens" in usage_data
+                        ):
                             reasoning_tokens = usage_data.get("reasoning_tokens", 0)
                     elif "model" in data_json and not model:
                         model = data_json["model"]

routstr/upstream/fake.py

@@ -1,265 +0,0 @@
-import asyncio
-import json
-import random
-from typing import AsyncIterator
-
-from fastapi import Request
-from fastapi.responses import Response, StreamingResponse
-
-from ..core.db import ApiKey, AsyncSession
-from ..payment.models import Architecture, Model, Pricing
-from .base import BaseUpstreamProvider
-
-
-class MockUpstreamProvider(BaseUpstreamProvider):
-    """Fack Mock Upstream provider specifically for Testing."""
-
-    provider_type = "mock"
-
-    async def forward_request(
-        self,
-        request: Request,
-        path: str,
-        headers: dict,
-        request_body: bytes | None,
-        key: ApiKey,
-        max_cost_for_model: int,
-        session: AsyncSession,
-        model_obj: Model,
-    ) -> Response | StreamingResponse:
-        if path.endswith("chat/completions"):
-            is_streaming = False
-            if request_body:
-                request_data = json.loads(request_body)
-                is_streaming = request_data.get("stream", False)
-
-            if is_streaming:
-
-                async def fake_streaming_response(
-                    chunk_size: int | None = None,
-                ) -> AsyncIterator[bytes]:
-                    suffix = random.randint(1000, 9999)
-                    req_id = f"gen-mock-stream-{suffix}"
-                    created = 1766138895
-                    model = "mock/gpt-420-mock"
-
-                    def make_chunk(
-                        delta: dict,
-                        finish_reason: str | None = None,
-                        usage: dict | None = None,
-                    ) -> bytes:
-                        chunk = {
-                            "id": req_id,
-                            "provider": "MockProvider",
-                            "model": model,
-                            "object": "chat.completion.chunk",
-                            "created": created,
-                            "choices": [
-                                {
-                                    "index": 0,
-                                    "delta": delta,
-                                    "finish_reason": finish_reason,
-                                    "native_finish_reason": "completed"
-                                    if finish_reason
-                                    else None,
-                                    "logprobs": None,
-                                }
-                            ],
-                        }
-                        if usage:
-                            chunk["usage"] = usage
-                        return f"data: {json.dumps(chunk)}\n\n".encode()
-
-                    # 1. Initial chunk
-                    yield make_chunk({"role": "assistant", "content": ""})
-                    await asyncio.sleep(0.02)
-
-                    # 2. Reasoning chunks
-                    reasoning_tokens = ["Mock", " reason", "ing", "..."]
-                    for token in reasoning_tokens:
-                        delta = {
-                            "role": "assistant",
-                            "content": "",
-                            "reasoning": token,
-                            "reasoning_details": [
-                                {
-                                    "type": "reasoning.summary",
-                                    "summary": token,
-                                    "format": "openai-responses-v1",
-                                    "index": 0,
-                                }
-                            ],
-                        }
-                        yield make_chunk(delta)
-                        await asyncio.sleep(0.03)
-
-                    # 3. Content chunks
-                    content_tokens = ["This", " is", " a", " mock", " stream", "."]
-                    for token in content_tokens:
-                        yield make_chunk({"role": "assistant", "content": token})
-                        await asyncio.sleep(0.03)
-
-                    # 4. Finish chunk
-                    yield make_chunk(
-                        {"role": "assistant", "content": ""}, finish_reason="stop"
-                    )
-
-                    # 5. Usage chunk
-                    usage_data = {
-                        "prompt_tokens": 10,
-                        "completion_tokens": 20,
-                        "total_tokens": 30,
-                        "cost": 0.001,
-                        "is_byok": False,
-                        "prompt_tokens_details": {
-                            "cached_tokens": 0,
-                            "audio_tokens": 0,
-                            "video_tokens": 0,
-                        },
-                        "cost_details": {
-                            "upstream_inference_cost": None,
-                            "upstream_inference_prompt_cost": 0,
-                            "upstream_inference_completions_cost": 0.001,
-                        },
-                        "completion_tokens_details": {
-                            "reasoning_tokens": 10,
-                            "image_tokens": 0,
-                        },
-                    }
-
-                    usage_chunk = {
-                        "id": req_id,
-                        "provider": "MockProvider",
-                        "model": model,
-                        "object": "chat.completion.chunk",
-                        "created": created,
-                        "choices": [
-                            {
-                                "index": 0,
-                                "delta": {"role": "assistant", "content": ""},
-                                "finish_reason": None,
-                                "native_finish_reason": None,
-                                "logprobs": None,
-                            }
-                        ],
-                        "usage": usage_data,
-                    }
-                    yield f"data: {json.dumps(usage_chunk)}\n\n".encode()
-
-                    # 6. DONE
-                    yield b"data: [DONE]\n\n"
-
-                    # 7. Cost
-                    cost_chunk = {
-                        "cost": {
-                            "base_msats": 0,
-                            "input_msats": 2,
-                            "output_msats": 10,
-                            "total_msats": 12,
-                        }
-                    }
-                    yield f"data: {json.dumps(cost_chunk)}\n\n".encode()
-
-                return StreamingResponse(
-                    fake_streaming_response(),
-                    200,
-                )
-
-            else:
-                suffix = random.randint(1000, 9999)
-                content_dict = {
-                    "id": f"gen-mock-{suffix}",
-                    "provider": "MockProvider",
-                    "model": "mock/gpt-5-mini",
-                    "object": "chat.completion",
-                    "created": 1766138655,
-                    "choices": [
-                        {
-                            "logprobs": None,
-                            "finish_reason": "length",
-                            "native_finish_reason": "max_output_tokens",
-                            "index": 0,
-                            "message": {
-                                "role": "assistant",
-                                "content": f"Mock Content {suffix}",
-                                "refusal": None,
-                                "reasoning": f"Mock Reasoning {suffix}",
-                                "reasoning_details": [
-                                    {
-                                        "format": "openai-responses-v1",
-                                        "index": 0,
-                                        "type": "reasoning.summary",
-                                        "summary": f"Mock Summary {suffix}",
-                                    },
-                                    {
-                                        "id": f"rs_mock_{suffix}",
-                                        "format": "openai-responses-v1",
-                                        "index": 0,
-                                        "type": "reasoning.encrypted",
-                                        "data": "mock_encrypted_data",
-                                    },
-                                ],
-                            },
-                        }
-                    ],
-                    "usage": {
-                        "prompt_tokens": 10,
-                        "completion_tokens": 10,
-                        "total_tokens": 20,
-                        "cost": 0,
-                        "is_byok": False,
-                        "prompt_tokens_details": {
-                            "cached_tokens": 0,
-                            "audio_tokens": 0,
-                            "video_tokens": 0,
-                        },
-                        "cost_details": {
-                            "upstream_inference_cost": None,
-                            "upstream_inference_prompt_cost": 0,
-                            "upstream_inference_completions_cost": 0,
-                        },
-                        "completion_tokens_details": {
-                            "reasoning_tokens": 5,
-                            "image_tokens": 0,
-                        },
-                    },
-                    "cost": {
-                        "base_msats": 0,
-                        "input_msats": 0,
-                        "output_msats": 0,
-                        "total_msats": 0,
-                    },
-                }
-                return Response(json.dumps(content_dict).encode(), 200)
-
-        elif path.endswith("embeddings"):
-            raise NotImplementedError
-        elif path.endswith("responses"):
-            raise NotImplementedError
-        else:
-            raise NotImplementedError
-
-    async def fetch_models(self) -> list[Model]:
-        return [
-            Model(
-                id="mock/gpt-420-mock",
-                name="mock/gpt-420-mock",
-                created=0,
-                description="mock model for testing",
-                context_length=8192,
-                architecture=Architecture(
-                    modality="text",
-                    input_modalities=["text"],
-                    output_modalities=["text"],
-                    tokenizer="",
-                    instruct_type=None,
-                ),
-                pricing=Pricing(prompt=0.01, completion=0.01),
-            ),
-        ]
-
-    def transform_model_name(self, model_id: str) -> str:
-        return "fake-model"
-
-    async def get_balance(self) -> float | None:
-        return 420.69

routstr/upstream/gemini.py

@@ -187,11 +187,44 @@ class GeminiUpstreamProvider(BaseUpstreamProvider):
                 )
 
                 async def stream_with_cost() -> AsyncGenerator[bytes, None]:
+                    payment_finalized = False
+
+                    async def finalize_payment() -> None:
+                        nonlocal payment_finalized
+                        if payment_finalized:
+                            return
+                        from ..auth import adjust_payment_for_tokens
+                        from ..core.db import create_session
+
+                        async with create_session() as new_session:
+                            fresh_key = await new_session.get(
+                                key.__class__, key.hashed_key
+                            )
+                            if fresh_key:
+                                try:
+                                    await adjust_payment_for_tokens(
+                                        fresh_key,
+                                        {
+                                            "model": model_obj.id,
+                                            "usage": final_usage_data,
+                                        },
+                                        new_session,
+                                        max_cost_for_model,
+                                    )
+                                    payment_finalized = True
+                                except Exception as cost_error:
+                                    logger.error(
+                                        "Error finalizing Gemini streaming payment in fallback",
+                                        extra={
+                                            "error": str(cost_error),
+                                            "key_hash": key.hashed_key[:8] + "...",
+                                        },
+                                    )
+
                     try:
                         async for chunk in response_generator:
                             sse_data = f"data: {json.dumps(chunk)}\n\n"
                             yield sse_data.encode()
-
                     except Exception as e:
                         logger.error(
                             "Error in Gemini streaming response",
@@ -202,6 +235,9 @@ class GeminiUpstreamProvider(BaseUpstreamProvider):
                             },
                         )
                         raise
+                    finally:
+                        if not payment_finalized:
+                            await finalize_payment()
 
                 return StreamingResponse(
                     stream_with_cost(),

routstr/upstream/helpers.py

@@ -102,6 +102,8 @@ async def get_all_models_with_overrides(
             )
             for row in override_rows
             if row.upstream_provider_id is not None
+            and row.upstream_provider_id in providers_by_id
+            and providers_by_id[row.upstream_provider_id].enabled
         }
 
     all_models: dict[str, Model] = {}
@@ -218,14 +220,6 @@ async def init_upstreams() -> list[BaseUpstreamProvider]:
         results = await asyncio.gather(*tasks)
         upstreams = [p for p in results if p is not None]
 
-        if "https://testnut.cashu.space" in settings.cashu_mints:
-            from .fake import MockUpstreamProvider
-
-            mock_provider = MockUpstreamProvider("mock", "mock")
-            await mock_provider.refresh_models_cache()
-            upstreams.append(mock_provider)
-            logger.info("Initialized MockUpstreamProvider for testnut mint")
-
         return upstreams
 
 
@@ -242,7 +236,7 @@ async def _seed_providers_from_settings(
     from . import upstream_provider_classes
 
     providers_to_add: list[UpstreamProviderRow] = []
-    seeded_base_urls: set[str] = set()
+    seeded_provider_keys: set[tuple[str, str]] = set()
 
     provider_classes_by_type = {
         cls.provider_type: cls
@@ -267,7 +261,8 @@ async def _seed_providers_from_settings(
                 base_url = provider_class.default_base_url  # type: ignore[attr-defined]
                 result = await session.exec(
                     select(UpstreamProviderRow).where(
-                        UpstreamProviderRow.base_url == base_url
+                        UpstreamProviderRow.base_url == base_url,
+                        UpstreamProviderRow.api_key == api_key,
                     )
                 )
                 if not result.first():
@@ -279,13 +274,15 @@ async def _seed_providers_from_settings(
                             enabled=True,
                         )
                     )
-                    seeded_base_urls.add(base_url)
+                    seeded_provider_keys.add((base_url, api_key))
 
     ollama_base_url = os.environ.get("OLLAMA_BASE_URL")
     if ollama_base_url:
+        ollama_api_key = os.environ.get("OLLAMA_API_KEY", "")
         result = await session.exec(
             select(UpstreamProviderRow).where(
-                UpstreamProviderRow.base_url == ollama_base_url
+                UpstreamProviderRow.base_url == ollama_base_url,
+                UpstreamProviderRow.api_key == ollama_api_key,
             )
         )
         if not result.first():
@@ -293,18 +290,20 @@ async def _seed_providers_from_settings(
                 UpstreamProviderRow(
                     provider_type="ollama",
                     base_url=ollama_base_url,
-                    api_key=os.environ.get("OLLAMA_API_KEY", ""),
+                    api_key=ollama_api_key,
                     enabled=True,
                 )
             )
-            seeded_base_urls.add(ollama_base_url)
+            seeded_provider_keys.add((ollama_base_url, ollama_api_key))
 
     if settings.chat_completions_api_version and settings.upstream_base_url:
         base_url = settings.upstream_base_url
-        if base_url not in seeded_base_urls:
+        api_key = settings.upstream_api_key
+        if (base_url, api_key) not in seeded_provider_keys:
             result = await session.exec(
                 select(UpstreamProviderRow).where(
-                    UpstreamProviderRow.base_url == base_url
+                    UpstreamProviderRow.base_url == base_url,
+                    UpstreamProviderRow.api_key == api_key,
                 )
             )
             if not result.first():
@@ -312,19 +311,21 @@ async def _seed_providers_from_settings(
                     UpstreamProviderRow(
                         provider_type="azure",
                         base_url=base_url,
-                        api_key=settings.upstream_api_key,
+                        api_key=api_key,
                         api_version=settings.chat_completions_api_version,
                         enabled=True,
                     )
                 )
-                seeded_base_urls.add(base_url)
+                seeded_provider_keys.add((base_url, api_key))
 
     if settings.upstream_base_url and settings.upstream_api_key:
         base_url = settings.upstream_base_url
-        if base_url not in seeded_base_urls:
+        api_key = settings.upstream_api_key
+        if (base_url, api_key) not in seeded_provider_keys:
             result = await session.exec(
                 select(UpstreamProviderRow).where(
-                    UpstreamProviderRow.base_url == base_url
+                    UpstreamProviderRow.base_url == base_url,
+                    UpstreamProviderRow.api_key == api_key,
                 )
             )
             if not result.first():
@@ -332,11 +333,11 @@ async def _seed_providers_from_settings(
                     UpstreamProviderRow(
                         provider_type="custom",
                         base_url=base_url,
-                        api_key=settings.upstream_api_key,
+                        api_key=api_key,
                         enabled=True,
                     )
                 )
-                seeded_base_urls.add(base_url)
+                seeded_provider_keys.add((base_url, api_key))
 
     for provider in providers_to_add:
         session.add(provider)

routstr/upstream/ppqai.py

@@ -196,6 +196,37 @@ class PPQAIUpstreamProvider(BaseUpstreamProvider):
             )
             return []
 
+    async def on_upstream_error_redirect(
+        self, status_code: int, error_message: str
+    ) -> None:
+        if "insufficient balance" in error_message.lower():
+            logger.warning(
+                f"Disabling PPQ.AI provider ({self.base_url}) due to insufficient balance",
+                extra={"error": error_message},
+            )
+            from sqlmodel import select
+
+            from ..core.db import UpstreamProviderRow, create_session
+
+            async with create_session() as session:
+                statement = select(UpstreamProviderRow).where(
+                    UpstreamProviderRow.base_url == self.base_url,
+                    UpstreamProviderRow.api_key == self.api_key,
+                )
+                result = await session.exec(statement)
+                provider = result.first()
+
+                if provider:
+                    provider.enabled = False
+                    session.add(provider)
+                    await session.commit()
+
+                    # Trigger re-initialization of providers
+                    # Import here to avoid circular dependency
+                    from ..proxy import reinitialize_upstreams
+
+                    await reinitialize_upstreams()
+
     async def create_account(self) -> dict[str, object]:
         """Create a new PPQ.AI account.
 

routstr/wallet.py

@@ -81,7 +81,7 @@ async def swap_to_primary_mint(
         amount_msat = token_amount
     else:
         raise ValueError("Invalid unit")
-    estimated_fee_sat = math.ceil(max(amount_msat // 1000 * 0.01, 2))
+    estimated_fee_sat = math.ceil(max(amount_msat // 1000 * 0.01, 2)) + 1
     amount_msat_after_fee = amount_msat - estimated_fee_sat * 1000
     primary_wallet = await get_wallet(settings.primary_mint, settings.primary_mint_unit)
 
@@ -313,14 +313,13 @@ async def periodic_payout() -> None:
         try:
             async with db.create_session() as session:
                 for mint_url in settings.cashu_mints:
-                    if mint_url == "https://testnut.cashu.space":
-                        continue
                     for unit in ["sat", "msat"]:
                         wallet = await get_wallet(mint_url, unit)
                         proofs = get_proofs_per_mint_and_unit(
                             wallet, mint_url, unit, not_reserved=True
                         )
                         proofs = await slow_filter_spend_proofs(proofs, wallet)
+                        await asyncio.sleep(5)
                         user_balance = await db.balances_for_mint_and_unit(
                             session, mint_url, unit
                         )
@@ -346,8 +345,6 @@ async def periodic_payout() -> None:
                                     "amount_received": amount_received,
                                 },
                             )
-
-                        await asyncio.sleep(5)
         except Exception as e:
             logger.error(
                 f"Error sending payout: {type(e).__name__}",

scripts/reproduce_bug_staging.py

@@ -0,0 +1,101 @@
+import asyncio
+
+import httpx
+
+BASE_URL = input("Enter routstr URL: ")
+API_KEY = input("Enter key or token: ")
+
+
+async def get_balance(client: httpx.AsyncClient) -> int:
+    response = await client.get("/v1/balance/info")
+    response.raise_for_status()
+    data = response.json()
+    print(f"Current Balance Info: {data}")
+    return data.get("reserved", 0)
+
+
+async def reproduce() -> None:
+    headers = {"Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"}
+
+    async with httpx.AsyncClient(
+        base_url=BASE_URL, headers=headers, timeout=30.0
+    ) as client:
+        print("Checking initial balance...")
+        try:
+            initial_reserved = await get_balance(client)
+        except Exception as e:
+            print(f"Failed to get balance: {e}")
+            return
+
+        print("\nStarting streaming request...")
+        try:
+            # Create a separate client for the stream so we can close it independently if needed,
+            # but usually just breaking the loop and exiting the context manager is enough.
+            # However, to be sure we simulate a harsh disconnect, we can just cancel the task or close the client.
+
+            async with client.stream(
+                "POST",
+                "/v1/chat/completions",
+                json={
+                    "model": "gpt-5-nano",
+                    "messages": [
+                        {
+                            "role": "user",
+                            "content": "Write a long poem about the ocean.",
+                        }
+                    ],
+                    "stream": True,
+                },
+            ) as response:
+                print(f"Stream status: {response.status_code}")
+                if response.status_code != 200:
+                    err_bytes = await response.aread()
+                    try:
+                        err_str = err_bytes.decode()
+                    except Exception:
+                        err_str = repr(err_bytes)
+                    print(f"Error: {err_str}")
+                    return
+
+                print("Stream started. Reading a few chunks...")
+                count = 0
+                async for chunk in response.aiter_bytes():
+                    print(f"Received chunk: {len(chunk)} bytes")
+                    count += 1
+                    if count >= 3:
+                        print("Simulating client disconnect (breaking stream)...")
+                        break
+
+        except Exception as e:
+            print(f"Stream interrupted (expected): {e}")
+
+        # Wait a bit for the server to realize we disconnected (though with asyncio it might be immediate or depend on keepalive)
+        print("\nWaiting for server to process disconnect...")
+        await asyncio.sleep(21)
+
+        print("\nChecking final balance...")
+        try:
+            final_reserved = await get_balance(client)
+        except Exception:
+            # Retry once if connection was closed
+            async with httpx.AsyncClient(
+                base_url=BASE_URL, headers=headers, timeout=30.0
+            ) as new_client:
+                final_reserved = await get_balance(new_client)
+
+        if final_reserved > initial_reserved:
+            print(
+                f"\n[FAIL] Bug reproduced! Reserved balance increased: {initial_reserved} -> {final_reserved}"
+            )
+            print(f"Accumulated reserved balance: {final_reserved - initial_reserved}")
+        else:
+            print(
+                f"\n[PASS] Reserved balance released correctly: {initial_reserved} -> {final_reserved}"
+            )
+
+
+if __name__ == "__main__":
+    try:
+        asyncio.run(reproduce())
+    except KeyboardInterrupt:
+        pass

tests/integration/test_child_keys.py

@@ -0,0 +1,141 @@
+import secrets
+from typing import Any
+
+import pytest
+from fastapi import HTTPException
+from sqlmodel.ext.asyncio.session import AsyncSession
+
+from routstr.auth import adjust_payment_for_tokens, pay_for_request
+from routstr.balance import ChildKeyRequest, create_child_key
+from routstr.core.db import ApiKey
+from routstr.core.settings import settings
+
+
+@pytest.mark.asyncio
+async def test_child_key_flow(integration_session: AsyncSession) -> None:
+    # 1. Create a parent key with balance
+    parent_raw = "parent_test_key_" + secrets.token_hex(4)
+    parent_key = ApiKey(
+        hashed_key=parent_raw,
+        balance=10000,  # 10 sats
+    )
+    integration_session.add(parent_key)
+    await integration_session.commit()
+    await integration_session.refresh(parent_key)
+
+    # Mock settings
+    settings.child_key_cost = 1000  # 1 sat
+
+    # 2. Call create_child_key
+    result = await create_child_key(
+        ChildKeyRequest(count=1), parent_key, integration_session
+    )
+
+    assert "api_keys" in result
+    assert result["cost_msats"] == 1000
+    assert result["parent_balance"] == 9000
+
+    child_key_raw = result["api_keys"][0][3:]  # remove sk-
+
+    # 3. Verify child key exists in DB
+    child_key_db = await integration_session.get(ApiKey, child_key_raw)
+    assert child_key_db is not None
+    assert child_key_db.parent_key_hash == parent_key.hashed_key
+    assert child_key_db.balance == 0
+
+    # 4. Test payment with child key
+    cost = 500
+    await pay_for_request(child_key_db, cost, integration_session)
+
+    # Refresh keys
+    await integration_session.refresh(parent_key)
+    await integration_session.refresh(child_key_db)
+
+    # Parent should be charged
+    assert parent_key.reserved_balance == 500
+    assert parent_key.total_requests == 1
+
+    # Child should have total_requests incremented
+    assert child_key_db.total_requests == 1
+
+    # 5. Test adjustment
+    response_data = {"model": "test-model", "usage": {"total_tokens": 10}}
+
+    # Mock calculate_cost
+    import routstr.auth
+    from routstr.payment.cost_calculation import CostData
+
+    async def mock_calculate_cost(*args: Any, **kwargs: Any) -> CostData:
+        return CostData(
+            base_msats=0, input_msats=200, output_msats=200, total_msats=400
+        )
+
+    # Patch calculate_cost
+    original_calculate_cost = routstr.auth.calculate_cost
+    routstr.auth.calculate_cost = mock_calculate_cost
+
+    try:
+        adjustment = await adjust_payment_for_tokens(
+            child_key_db, response_data, integration_session, 500
+        )
+        assert adjustment["total_msats"] == 400
+
+        # Refresh keys
+        await integration_session.refresh(parent_key)
+        await integration_session.refresh(child_key_db)
+
+        # Parent should have updated balance and total_spent
+        assert parent_key.reserved_balance == 0
+        assert parent_key.balance == 9000 - 400
+        assert (
+            parent_key.total_spent == 1400
+        )  # 1000 for child key creation + 400 for request
+
+        # Child should also have total_spent updated
+        assert child_key_db.total_spent == 400
+
+    finally:
+        routstr.auth.calculate_cost = original_calculate_cost
+
+
+@pytest.mark.asyncio
+async def test_child_key_insufficient_balance(
+    integration_session: AsyncSession,
+) -> None:
+    parent_key = ApiKey(
+        hashed_key="poor_parent_" + secrets.token_hex(4),
+        balance=500,
+    )
+    integration_session.add(parent_key)
+    await integration_session.commit()
+    await integration_session.refresh(parent_key)
+
+    settings.child_key_cost = 1000
+
+    with pytest.raises(HTTPException) as exc:
+        await create_child_key(
+            ChildKeyRequest(count=1), parent_key, integration_session
+        )
+    assert exc.value.status_code == 402
+
+
+@pytest.mark.asyncio
+async def test_child_key_cannot_create_child(integration_session: AsyncSession) -> None:
+    parent_key = ApiKey(
+        hashed_key="parent_" + secrets.token_hex(4),
+        balance=10000,
+    )
+    child_key = ApiKey(
+        hashed_key="child_" + secrets.token_hex(4),
+        balance=0,
+        parent_key_hash=parent_key.hashed_key,
+    )
+    integration_session.add(parent_key)
+    integration_session.add(child_key)
+    await integration_session.commit()
+    await integration_session.refresh(child_key)
+
+    with pytest.raises(HTTPException) as exc:
+        await create_child_key(ChildKeyRequest(count=1), child_key, integration_session)
+    assert exc.value.status_code == 400
+    assert "Cannot create a child key for another child key" in str(exc.value.detail)

tests/integration/test_example.py

@@ -10,7 +10,6 @@ from httpx import AsyncClient
 
 from .utils import (
     CashuTokenGenerator,
-    PerformanceValidator,
     ResponseValidator,
 )
 
@@ -159,30 +158,7 @@ async def test_error_handling(
     assert response.status_code == 401
 
 
-@pytest.mark.integration
-@pytest.mark.asyncio
-async def test_performance_requirements(integration_client: AsyncClient) -> None:
-    """Test that endpoints meet performance requirements"""
 
-    validator = PerformanceValidator()
-
-    # Test info endpoint performance
-    for i in range(50):
-        start = validator.start_timing("info_endpoint")
-        response = await integration_client.get("/")
-        validator.end_timing("info_endpoint", start)
-        assert response.status_code == 200
-
-    # Validate 95th percentile is under 500ms
-    result = validator.validate_response_time(
-        "info_endpoint", max_duration=0.5, percentile=0.95
-    )
-
-    assert result["valid"], (
-        f"Performance requirement failed: "
-        f"95th percentile was {result['percentile_time']:.3f}s "
-        f"(required < {result['max_allowed']}s)"
-    )
 
 
 @pytest.mark.integration

tests/integration/test_general_info_endpoints.py

@@ -267,10 +267,9 @@ async def test_admin_endpoint_unauthenticated(
     """Test GET /admin/ endpoint redirects to /"""
     await db_snapshot.capture()
 
-    response = await integration_client.get("/admin/")
+    response = await integration_client.get("/admin/api/settings")
 
-    assert response.status_code == 307
-    assert response.headers.get("location") == "/"
+    assert response.status_code == 403
 
     diff = await db_snapshot.diff()
     assert len(diff["api_keys"]["added"]) == 0

tests/integration/test_key_logic.py

@@ -0,0 +1,142 @@
+import time
+from datetime import datetime, timedelta
+
+import pytest
+from sqlmodel import select
+from sqlmodel.ext.asyncio.session import AsyncSession
+
+from routstr.auth import pay_for_request
+from routstr.core.db import ApiKey
+
+
+@pytest.mark.asyncio
+async def test_key_validity_date(integration_session: AsyncSession) -> None:
+    # 1. Create a key that is expired
+    expired_time = int(time.time()) - 3600
+    key = ApiKey(hashed_key="expired_key", balance=1000, validity_date=expired_time)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    # 2. Try to pay for a request - should fail
+    with pytest.raises(Exception) as excinfo:
+        await pay_for_request(key, 100, integration_session)
+    assert "expired" in str(excinfo.value).lower()
+
+
+@pytest.mark.asyncio
+async def test_key_balance_limit(integration_session: AsyncSession) -> None:
+    # 1. Create a key with a balance limit
+    key = ApiKey(
+        hashed_key="limited_key", balance=10000, balance_limit=500, total_spent=450
+    )
+    integration_session.add(key)
+    await integration_session.commit()
+
+    # 2. Try to pay for a request that exceeds the limit
+    with pytest.raises(Exception) as excinfo:
+        await pay_for_request(key, 100, integration_session)
+    assert "limit exceeded" in str(excinfo.value).lower()
+
+    # 3. Try to pay for a request that fits
+    await pay_for_request(key, 50, integration_session)
+    await integration_session.refresh(key)
+    # Note: total_spent is updated in adjust_payment_for_tokens,
+    # but pay_for_request checks it.
+    # In our current logic, pay_for_request checks (total_spent + cost) > balance_limit.
+
+
+@pytest.mark.asyncio
+async def test_key_daily_reset_policy(integration_session: AsyncSession) -> None:
+    # 1. Create a key with a daily reset policy and old reset date
+    yesterday = int((datetime.now() - timedelta(days=1)).timestamp())
+    key = ApiKey(
+        hashed_key="daily_reset_key",
+        balance=10000,
+        balance_limit=1000,
+        balance_limit_reset="daily",
+        balance_limit_reset_date=yesterday,
+        total_spent=900,
+    )
+    integration_session.add(key)
+    await integration_session.commit()
+
+    # 2. Pay for a request - should trigger reset first because it's a new day
+    # Request is 200, total_spent is 900. 900+200 > 1000,
+    # but reset should happen making total_spent 0, then 0+200 < 1000.
+    await pay_for_request(key, 200, integration_session)
+
+    await integration_session.refresh(key)
+    assert key.total_spent == 0  # Reset in pay_for_request happens before charging
+    # Wait, the charging logic in pay_for_request increments parent/billing_key's total_requests,
+    # but total_spent is updated in adjust_payment_for_tokens.
+    # However, the reset logic sets total_spent to 0.
+    assert key.balance_limit_reset_date is not None
+    assert key.balance_limit_reset_date > yesterday
+
+
+@pytest.mark.asyncio
+async def test_periodic_key_reset_job(integration_session: AsyncSession) -> None:
+    # 1. Create multiple keys needing reset
+    yesterday = int((datetime.now() - timedelta(days=1)).timestamp())
+    key1 = ApiKey(
+        hashed_key="job_reset_key_1",
+        balance=1000,
+        balance_limit=1000,
+        balance_limit_reset="daily",
+        balance_limit_reset_date=yesterday,
+        total_spent=500,
+    )
+    key2 = ApiKey(
+        hashed_key="job_reset_key_2",
+        balance=1000,
+        balance_limit=1000,
+        balance_limit_reset="daily",
+        balance_limit_reset_date=yesterday,
+        total_spent=800,
+    )
+    integration_session.add(key1)
+    integration_session.add(key2)
+    await integration_session.commit()
+
+    # 2. Run the periodic reset logic manually (mocking the background task loop)
+    # We can't easily run the actual loop because it has a sleep,
+    # but we can test the logic inside.
+
+    # Implementation of periodic_key_reset logic for testing:
+    stmt = select(ApiKey).where(ApiKey.balance_limit_reset != None)  # noqa: E711
+    keys = (await integration_session.exec(stmt)).all()
+    now = int(time.time())
+    for k in keys:
+        if k.hashed_key in ["job_reset_key_1", "job_reset_key_2"]:
+            k.total_spent = 0
+            k.balance_limit_reset_date = now
+            integration_session.add(k)
+    await integration_session.commit()
+
+    # 3. Verify resets
+    await integration_session.refresh(key1)
+    await integration_session.refresh(key2)
+    assert key1.total_spent == 0
+    assert key2.total_spent == 0
+
+
+@pytest.mark.asyncio
+async def test_refund_does_not_delete_key(integration_session: AsyncSession) -> None:
+    # This requires mocking the router call or testing the logic in balance.py
+    from routstr.balance import ApiKey
+
+    key = ApiKey(hashed_key="refund_test_key", balance=1000, reserved_balance=100)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    # Logic from refund_wallet_endpoint:
+    key.balance = 0
+    key.reserved_balance = 0
+    integration_session.add(key)
+    await integration_session.commit()
+
+    # Verify key still exists
+    fetched_key = await integration_session.get(ApiKey, "refund_test_key")
+    assert fetched_key is not None
+    assert fetched_key.balance == 0
+    assert fetched_key.reserved_balance == 0

tests/integration/test_performance_load.py

@@ -9,6 +9,7 @@ import gc
 import statistics
 import time
 from typing import Any, Dict, List
+from unittest.mock import patch
 
 import psutil
 import pytest
@@ -105,42 +106,46 @@ class TestPerformanceBaseline:
             ("GET", "/v1/wallet/info", authenticated_client, None),
         ]
 
-        # Warm up
-        for _ in range(10):
-            await integration_client.get("/")
+        # Enable provider discovery for this test
+        with patch(
+            "routstr.core.settings.settings.providers_refresh_interval_seconds", 300
+        ):
+            # Warm up
+            for _ in range(10):
+                await integration_client.get("/")
 
-        # Test each endpoint
-        for method, path, client, data in endpoints:
-            response_times = []
+            # Test each endpoint
+            for method, path, client, data in endpoints:
+                response_times = []
 
-            for i in range(100):
-                start = time.time()
+                for i in range(100):
+                    start = time.time()
 
-                if method == "GET":
-                    response = await client.get(path)
-                else:
-                    response = await client.post(path, json=data)
+                    if method == "GET":
+                        response = await client.get(path)
+                    else:
+                        response = await client.post(path, json=data)
 
-                duration = time.time() - start
-                response_times.append(duration * 1000)  # Convert to ms
+                    duration = time.time() - start
+                    response_times.append(duration * 1000)  # Convert to ms
 
-                assert response.status_code in [200, 201]
+                    assert response.status_code in [200, 201]
 
-                if i % 10 == 0:
-                    metrics.record_system_metrics()
+                    if i % 10 == 0:
+                        metrics.record_system_metrics()
 
-            # Verify 95th percentile < 500ms
-            p95 = sorted(response_times)[int(len(response_times) * 0.95)]
-            assert p95 < 500, (
-                f"{method} {path} p95 response time {p95}ms exceeds 500ms limit"
-            )
+                # Verify 95th percentile < 500ms
+                p95 = sorted(response_times)[int(len(response_times) * 0.95)]
+                assert p95 < 500, (
+                    f"{method} {path} p95 response time {p95}ms exceeds 500ms limit"
+                )
 
-            print(f"\n{method} {path}:")
-            print(f"  Mean: {statistics.mean(response_times):.2f}ms")
-            print(f"  P95: {p95:.2f}ms")
-            print(
-                f"  P99: {sorted(response_times)[int(len(response_times) * 0.99)]:.2f}ms"
-            )
+                print(f"\n{method} {path}:")
+                print(f"  Mean: {statistics.mean(response_times):.2f}ms")
+                print(f"  P95: {p95:.2f}ms")
+                print(
+                    f"  P99: {sorted(response_times)[int(len(response_times) * 0.99)]:.2f}ms"
+                )
 
 
 @pytest.mark.integration

tests/integration/test_provider_management.py

@@ -3,15 +3,15 @@ Integration tests for provider management functionality.
 Tests GET /v1/providers/ endpoint for listing and managing providers.
 """
 
-from typing import Any
+from typing import Any, Generator
 from unittest.mock import patch
 
 import pytest
 from httpx import AsyncClient
 
-from routstr.discovery import _PROVIDERS_CACHE
+from routstr.nostr.discovery import _PROVIDERS_CACHE
 
-from .utils import PerformanceValidator, ResponseValidator
+from .utils import ResponseValidator
 
 
 @pytest.fixture(autouse=True)
@@ -19,6 +19,15 @@ def _clear_providers_cache() -> None:
     _PROVIDERS_CACHE.clear()
 
 
+@pytest.fixture(autouse=True)
+def _enable_provider_discovery() -> Generator[None, Any, Any]:
+    """Enable provider discovery for all tests in this module"""
+    with patch(
+        "routstr.core.settings.settings.providers_refresh_interval_seconds", 300
+    ):
+        yield
+
+
 @pytest.mark.integration
 @pytest.mark.asyncio
 async def test_providers_endpoint_default_response(
@@ -62,9 +71,10 @@ async def test_providers_endpoint_default_response(
     }
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
+        with patch("routstr.nostr.discovery.fetch_provider_health") as mock_fetch:
             # Configure mock to return appropriate responses
             mock_fetch.side_effect = lambda url: mock_fetch_responses.get(
                 url, {"status_code": 500, "json": {"error": "Unknown provider"}}
@@ -126,9 +136,10 @@ async def test_providers_endpoint_with_include_json(
     }
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
+        with patch("routstr.nostr.discovery.fetch_provider_health") as mock_fetch:
             mock_fetch.return_value = {
                 "status_code": 200,
                 "json": mock_provider_response,
@@ -200,9 +211,10 @@ async def test_providers_data_structure_validation(
     }
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
+        with patch("routstr.nostr.discovery.fetch_provider_health") as mock_fetch:
             mock_fetch.return_value = mock_health_response
 
             response = await integration_client.get("/v1/providers/?include_json=true")
@@ -247,7 +259,8 @@ async def test_providers_endpoint_no_providers_found(
     ]
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
         response = await integration_client.get("/v1/providers/")
 
@@ -308,10 +321,11 @@ async def test_providers_endpoint_offline_providers(
             }
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
         with patch(
-            "routstr.discovery.fetch_provider_health",
+            "routstr.nostr.discovery.fetch_provider_health",
             side_effect=mock_fetch_provider_health,
         ):
             response = await integration_client.get("/v1/providers/?include_json=true")
@@ -377,9 +391,10 @@ async def test_providers_endpoint_duplicate_urls(
     ]
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
+        with patch("routstr.nostr.discovery.fetch_provider_health") as mock_fetch:
             mock_fetch.return_value = {
                 "status_code": 200,
                 "endpoint": "root",
@@ -416,7 +431,8 @@ async def test_providers_endpoint_nostr_relay_failures(
         raise Exception("Connection to relay failed")
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", side_effect=failing_query
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        side_effect=failing_query,
     ):
         response = await integration_client.get("/v1/providers/")
 
@@ -454,9 +470,10 @@ async def test_providers_endpoint_malformed_urls(
     ]
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
+        with patch("routstr.nostr.discovery.fetch_provider_health") as mock_fetch:
             mock_fetch.return_value = {"status_code": 200, "json": {"status": "online"}}
 
             response = await integration_client.get("/v1/providers/")
@@ -486,9 +503,10 @@ async def test_providers_endpoint_response_format(
     ]
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
+        with patch("routstr.nostr.discovery.fetch_provider_health") as mock_fetch:
             mock_fetch.return_value = {"status_code": 200, "json": {"status": "online"}}
 
             # Test default format
@@ -518,46 +536,6 @@ async def test_providers_endpoint_response_format(
             assert isinstance(data_json["providers"], list)
 
 
-@pytest.mark.integration
-@pytest.mark.asyncio
-async def test_providers_endpoint_performance(integration_client: AsyncClient) -> None:
-    """Test providers endpoint meets performance requirements"""
-
-    # Mock quick responses to avoid network delays
-    mock_events: list[dict[str, Any]] = [
-        {
-            "id": f"event{i}",
-            "content": f"Provider: http://provider{i}.onion",
-            "created_at": 1234567890 + i,
-        }
-        for i in range(5)
-    ]
-
-    validator = PerformanceValidator()
-
-    with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
-    ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
-            mock_fetch.return_value = {"status_code": 200, "json": {"status": "online"}}
-
-            # Test multiple requests
-            for i in range(10):
-                start = validator.start_timing("providers_endpoint")
-                response = await integration_client.get("/v1/providers/")
-                validator.end_timing("providers_endpoint", start)
-
-                assert response.status_code == 200
-
-    # Validate performance (should be fast with mocked dependencies)
-    perf_result = validator.validate_response_time(
-        "providers_endpoint",
-        max_duration=2.0,  # Allow more time since it involves multiple operations
-        percentile=0.95,
-    )
-    assert perf_result["valid"], f"Performance requirement failed: {perf_result}"
-
-
 @pytest.mark.integration
 @pytest.mark.asyncio
 async def test_providers_endpoint_concurrent_requests(
@@ -576,9 +554,10 @@ async def test_providers_endpoint_concurrent_requests(
     ]
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
+        with patch("routstr.nostr.discovery.fetch_provider_health") as mock_fetch:
             mock_fetch.return_value = {"status_code": 200, "json": {"status": "online"}}
 
             # Create concurrent requests
@@ -618,9 +597,10 @@ async def test_providers_endpoint_parameter_validation(
     ]
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
+        with patch("routstr.nostr.discovery.fetch_provider_health") as mock_fetch:
             mock_fetch.return_value = {"status_code": 200, "json": {"status": "online"}}
 
             # Test various parameter values
@@ -670,9 +650,10 @@ async def test_no_database_changes_during_provider_operations(
     ]
 
     with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
+        "routstr.nostr.discovery.query_nostr_relay_for_providers",
+        return_value=mock_events,
     ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
+        with patch("routstr.nostr.discovery.fetch_provider_health") as mock_fetch:
             mock_fetch.return_value = {"status_code": 200, "json": {"status": "online"}}
 
             # Make multiple requests with different parameters

tests/integration/test_proxy_get_endpoints.py

@@ -18,7 +18,6 @@ from routstr.core.db import ApiKey
 
 from .utils import (
     ConcurrencyTester,
-    PerformanceValidator,
 )
 
 
@@ -551,39 +550,7 @@ async def test_proxy_get_concurrent_requests(
             assert response.status_code == 200
 
 
-@pytest.mark.integration
-@pytest.mark.asyncio
-async def test_proxy_get_performance_requirements(
-    integration_client: AsyncClient, authenticated_client: AsyncClient
-) -> None:
-    """Test that GET proxy requests meet performance requirements"""
 
-    validator = PerformanceValidator()
-
-    with patch("httpx.AsyncClient.request") as mock_request:
-        mock_response = AsyncMock()
-        mock_response.status_code = 200
-        mock_response.headers = {"content-type": "application/json"}
-        mock_response.json = MagicMock(return_value={"performance": "test"})
-        mock_response.text = '{"performance": "test"}'
-        mock_response.iter_bytes = AsyncMock(return_value=[b'{"performance": "test"}'])
-        mock_request.return_value = mock_response
-
-        # Test multiple requests for performance measurement
-        for i in range(20):
-            start = validator.start_timing("proxy_get")
-            response = await authenticated_client.get(f"/v1/perf-test-{i}")
-            validator.end_timing("proxy_get", start)
-
-            assert response.status_code == 200
-
-    # Validate performance requirements
-    perf_result = validator.validate_response_time(
-        "proxy_get",
-        max_duration=1.0,  # Should complete within 1 second
-        percentile=0.95,
-    )
-    assert perf_result["valid"], f"Performance requirement failed: {perf_result}"
 
 
 @pytest.mark.integration

tests/integration/test_proxy_post_endpoints.py

@@ -15,7 +15,6 @@ from httpx import ASGITransport, AsyncClient
 
 from .utils import (
     ConcurrencyTester,
-    PerformanceValidator,
 )
 
 
@@ -290,55 +289,7 @@ async def test_proxy_post_unauthorized_access(integration_client: AsyncClient) -
     assert response.status_code in [400, 401]
 
 
-@pytest.mark.integration
-@pytest.mark.asyncio
-async def test_proxy_post_performance(
-    integration_client: AsyncClient, authenticated_client: AsyncClient
-) -> None:
-    """Test POST endpoint performance requirements"""
 
-    test_payload = {
-        "model": "gpt-3.5-turbo",
-        "messages": [{"role": "user", "content": "Performance test"}],
-    }
-
-    validator = PerformanceValidator()
-
-    with patch("httpx.AsyncClient.send") as mock_send:
-        # Mock fast responses
-        async def mock_iter_bytes(*args: Any, **kwargs: Any) -> Any:
-            yield b'{"choices": [{"message": {"content": "Fast"}}], "usage": {"total_tokens": 5}}'
-
-        mock_response = AsyncMock()
-        mock_response.status_code = 200
-        mock_response.headers = {"content-type": "application/json"}
-        response_data = {
-            "choices": [{"message": {"content": "Fast"}}],
-            "usage": {"total_tokens": 5},
-        }
-        mock_response.text = json.dumps(response_data)
-        mock_response.json = AsyncMock(return_value=response_data)
-        mock_response.iter_bytes = mock_iter_bytes
-        mock_response.aiter_bytes = mock_iter_bytes
-        mock_send.return_value = mock_response
-
-        # Run multiple requests for performance measurement
-        for i in range(20):
-            start = validator.start_timing("proxy_post")
-            response = await authenticated_client.post(
-                "/v1/chat/completions", json=test_payload
-            )
-            validator.end_timing("proxy_post", start)
-
-            assert response.status_code == 200
-
-    # Validate performance
-    perf_result = validator.validate_response_time(
-        "proxy_post",
-        max_duration=1.5,  # Allow slightly more time for POST
-        percentile=0.95,
-    )
-    assert perf_result["valid"], f"Performance requirement failed: {perf_result}"
 
 
 @pytest.mark.integration

tests/integration/test_wallet_information.py

@@ -3,7 +3,7 @@ Integration tests for wallet information retrieval endpoints.
 Tests GET /v1/wallet/ and GET /v1/wallet/info endpoints with various scenarios.
 """
 
-import time
+
 from datetime import datetime, timedelta
 from typing import Any
 
@@ -367,30 +367,4 @@ async def test_wallet_info_with_special_characters_in_headers(
     # Note: Current implementation doesn't return refund_address in response
 
 
-@pytest.mark.integration
-@pytest.mark.asyncio
-@pytest.mark.slow
-async def test_wallet_endpoints_performance(authenticated_client: AsyncClient) -> None:
-    """Test wallet endpoints meet performance requirements"""
 
-    # Warm up
-    await authenticated_client.get("/v1/wallet/")
-
-    # Measure response times
-    response_times = []
-
-    for _ in range(50):
-        start_time = time.time()
-        response = await authenticated_client.get("/v1/wallet/")
-        end_time = time.time()
-
-        assert response.status_code == 200
-        response_times.append(end_time - start_time)
-
-    # Calculate statistics
-    avg_time = sum(response_times) / len(response_times)
-    max_time = max(response_times)
-
-    # Performance assertions
-    assert avg_time < 0.1  # Average should be under 100ms
-    assert max_time < 0.5  # No request should take more than 500ms

tests/integration/test_wallet_refund.py

@@ -65,9 +65,10 @@ async def test_full_balance_refund_returns_cashu_token(
     except Exception as e:
         pytest.fail(f"Invalid Cashu token format: {e}")
 
-    # Try to use the API key - should fail since it's been deleted
+    # Try to use the API key - should still work but have 0 balance
     response = await authenticated_client.get("/v1/wallet/")
-    assert response.status_code == 401
+    assert response.status_code == 200
+    assert response.json()["balance"] == 0
 
     # The refund token has been validated above by decoding it
     # The API key deletion has been verified by the 401 response
@@ -261,17 +262,16 @@ async def test_database_state_after_refund(
     response = await authenticated_client.post("/v1/wallet/refund")
     assert response.status_code == 200
 
-    # Verify key is deleted after refund
-    result = await integration_session.execute(
-        select(ApiKey).where(ApiKey.hashed_key == hashed_key)  # type: ignore[arg-type]
-    )
-    assert result.scalar_one_or_none() is None
+    # Refresh the key to get the updated balance from the database
+    await integration_session.refresh(key_before)
 
-    # Count total keys to ensure only the specific one was deleted
+    # Verify key balance is 0 after refund
+    assert key_before.balance == 0
+
+    # Count total keys to ensure it wasn't deleted
     result = await integration_session.execute(select(ApiKey))
     remaining_keys = result.scalars().all()
-    # Should have no keys left (assuming clean test environment)
-    assert len(remaining_keys) == 0
+    assert len(remaining_keys) == 1
 
 
 @pytest.mark.integration
@@ -388,9 +388,10 @@ async def test_refund_during_active_usage(
     # Refund should succeed
     assert refund_response.status_code == 200
 
-    # Further usage should fail
+    # Further usage should return 200 but with 0 balance
     response = await authenticated_client.get("/v1/wallet/")
-    assert response.status_code == 401
+    assert response.status_code == 200
+    assert response.json()["balance"] == 0
 
 
 @pytest.mark.integration
@@ -535,44 +536,3 @@ async def test_refund_with_expired_key(
         # Should still allow manual refund
         assert response.status_code == 200
         assert response.json()["recipient"] == "expired@ln.address"
-
-
-@pytest.mark.integration
-@pytest.mark.asyncio
-@pytest.mark.slow
-async def test_refund_performance(
-    integration_client: AsyncClient, testmint_wallet: Any
-) -> None:
-    """Test refund endpoint performance"""
-
-    import time
-
-    # Create multiple API keys
-    api_keys = []
-    for i in range(10):
-        token = await testmint_wallet.mint_tokens(100 + i)
-        # Use cashu token as Bearer auth to create API key
-        integration_client.headers["Authorization"] = f"Bearer {token}"
-        response = await integration_client.get("/v1/wallet/info")
-        assert response.status_code == 200
-        api_keys.append(response.json()["api_key"])
-
-    # Measure refund times
-    refund_times = []
-
-    for api_key in api_keys:
-        integration_client.headers["Authorization"] = f"Bearer {api_key}"
-
-        start_time = time.time()
-        response = await integration_client.post("/v1/wallet/refund")
-        end_time = time.time()
-
-        assert response.status_code == 200
-        refund_times.append(end_time - start_time)
-
-    # Performance assertions
-    avg_time = sum(refund_times) / len(refund_times)
-    max_time = max(refund_times)
-
-    assert avg_time < 0.5  # Average under 500ms
-    assert max_time < 1.0  # No refund takes more than 1 second

tests/unit/test_algorithm.py

@@ -10,7 +10,6 @@ os.environ["UPSTREAM_API_KEY"] = "test"
 from routstr.algorithm import (  # noqa: E402
     calculate_model_cost_score,
     get_provider_penalty,
-    should_prefer_model,
 )
 from routstr.payment.models import Architecture, Model, Pricing  # noqa: E402
 
@@ -100,100 +99,3 @@ def test_get_provider_penalty_openrouter() -> None:
     provider = create_test_provider("openrouter", "https://openrouter.ai/api/v1")
     penalty = get_provider_penalty(provider)
     assert penalty == 1.001
-
-
-def test_should_prefer_model_cheaper_wins() -> None:
-    """Test that cheaper model is preferred."""
-    cheap_model = create_test_model("cheap", prompt_price=0.001, completion_price=0.002)
-    expensive_model = create_test_model(
-        "expensive", prompt_price=0.03, completion_price=0.06
-    )
-
-    provider1 = create_test_provider("provider1")
-    provider2 = create_test_provider("provider2")
-
-    # Cheaper model should win
-    assert should_prefer_model(
-        cheap_model, provider1, expensive_model, provider2, "test-alias"
-    )
-
-    # More expensive model should not win
-    assert not should_prefer_model(
-        expensive_model, provider2, cheap_model, provider1, "test-alias"
-    )
-
-
-def test_should_prefer_model_exact_match_wins() -> None:
-    """Test that exact alias match beats cheaper price."""
-    # Make model IDs match the alias differently
-    exact_match = create_test_model(
-        "test-model", prompt_price=0.03, completion_price=0.06
-    )
-    no_match = create_test_model(
-        "other-model", prompt_price=0.001, completion_price=0.002
-    )
-
-    provider1 = create_test_provider("provider1")
-    provider2 = create_test_provider("provider2")
-
-    # Exact match should win even though it's more expensive
-    assert should_prefer_model(
-        exact_match, provider1, no_match, provider2, "test-model"
-    )
-
-
-def test_should_prefer_model_openrouter_slight_penalty() -> None:
-    """Test that OpenRouter has slight penalty compared to other providers."""
-    model1 = create_test_model("model1", prompt_price=0.001, completion_price=0.002)
-    model2 = create_test_model("model2", prompt_price=0.001, completion_price=0.002)
-
-    regular_provider = create_test_provider("regular", "http://provider.com")
-    openrouter_provider = create_test_provider(
-        "openrouter", "https://openrouter.ai/api/v1"
-    )
-
-    # Regular provider should be preferred over OpenRouter at same cost
-    assert should_prefer_model(
-        model1, regular_provider, model2, openrouter_provider, "test-alias"
-    )
-
-    # OpenRouter should not replace regular provider at same cost
-    assert not should_prefer_model(
-        model2, openrouter_provider, model1, regular_provider, "test-alias"
-    )
-
-
-def test_should_prefer_model_openrouter_can_win_if_cheaper() -> None:
-    """Test that OpenRouter can still win if significantly cheaper."""
-    cheap_model = create_test_model(
-        "cheap", prompt_price=0.0001, completion_price=0.0002
-    )
-    expensive_model = create_test_model(
-        "expensive", prompt_price=0.03, completion_price=0.06
-    )
-
-    regular_provider = create_test_provider("regular", "http://provider.com")
-    openrouter_provider = create_test_provider(
-        "openrouter", "https://openrouter.ai/api/v1"
-    )
-
-    # OpenRouter should win if it's much cheaper (even with penalty)
-    assert should_prefer_model(
-        cheap_model,
-        openrouter_provider,
-        expensive_model,
-        regular_provider,
-        "test-alias",
-    )
-
-
-def test_should_prefer_model_same_cost_first_wins() -> None:
-    """Test that when costs are identical, current model is kept."""
-    model1 = create_test_model("model1", prompt_price=0.001, completion_price=0.002)
-    model2 = create_test_model("model2", prompt_price=0.001, completion_price=0.002)
-
-    provider1 = create_test_provider("provider1")
-    provider2 = create_test_provider("provider2")
-
-    # When costs are equal, should not replace
-    assert not should_prefer_model(model2, provider2, model1, provider1, "test-alias")

tests/unit/test_fee_consistency.py

@@ -1,155 +0,0 @@
-"""Unit tests for model row payload conversion.
-
-This module tests that _model_to_row_payload correctly serializes model data
-for database storage. Pricing is stored as-is without fee application.
-Fees are now applied per-provider when reading from the database.
-
-Key behaviors tested:
-1. Pricing is stored as-is without fee application
-2. All model fields are correctly serialized to JSON
-3. Optional fields are handled correctly (None values)
-4. Pricing structure is preserved
-5. Original model objects are not mutated
-"""
-
-import json
-import os
-
-import pytest
-
-# Set required env vars before importing
-os.environ["UPSTREAM_BASE_URL"] = "http://test"
-os.environ["UPSTREAM_API_KEY"] = "test"
-
-from routstr.payment.models import (  # noqa: E402
-    Architecture,
-    Model,
-    Pricing,
-    _model_to_row_payload,
-)
-
-
-@pytest.fixture
-def base_architecture() -> Architecture:
-    """Provide standard architecture for test models."""
-    return Architecture(
-        modality="text",
-        input_modalities=["text"],
-        output_modalities=["text"],
-        tokenizer="gpt",
-        instruct_type="chat",
-    )
-
-
-@pytest.fixture
-def standard_pricing() -> Pricing:
-    """Provide standard USD pricing with known values for testing."""
-    return Pricing(
-        prompt=0.001,
-        completion=0.002,
-        request=0.01,
-        image=0.05,
-        web_search=0.03,
-        internal_reasoning=0.015,
-        max_prompt_cost=10.0,
-        max_completion_cost=20.0,
-        max_cost=30.0,
-    )
-
-
-@pytest.fixture
-def standard_model(base_architecture: Architecture, standard_pricing: Pricing) -> Model:
-    """Create a standard test model with known pricing."""
-    return Model(
-        id="test-model-standard",
-        name="Test Model Standard",
-        created=1234567890,
-        description="A standard test model",
-        context_length=8192,
-        architecture=base_architecture,
-        pricing=standard_pricing,
-    )
-
-
-def test_pricing_stored_without_fees(standard_model: Model) -> None:
-    """Verify pricing is stored as-is without any fee application."""
-    payload = _model_to_row_payload(standard_model)
-    pricing_str = payload["pricing"]
-    assert isinstance(pricing_str, str)
-    pricing = json.loads(pricing_str)
-
-    assert pricing["prompt"] == pytest.approx(0.001, rel=1e-9)
-    assert pricing["completion"] == pytest.approx(0.002, rel=1e-9)
-    assert pricing["request"] == pytest.approx(0.01, rel=1e-9)
-    assert pricing["image"] == pytest.approx(0.05, rel=1e-9)
-    assert pricing["web_search"] == pytest.approx(0.03, rel=1e-9)
-    assert pricing["internal_reasoning"] == pytest.approx(0.015, rel=1e-9)
-    assert pricing["max_prompt_cost"] == pytest.approx(10.0, rel=1e-9)
-    assert pricing["max_completion_cost"] == pytest.approx(20.0, rel=1e-9)
-    assert pricing["max_cost"] == pytest.approx(30.0, rel=1e-9)
-
-
-def test_zero_value_pricing_fields(base_architecture: Architecture) -> None:
-    """Verify that zero-value pricing fields are stored correctly."""
-    zero_pricing = Pricing(
-        prompt=0.0,
-        completion=0.0,
-        request=0.0,
-        image=0.0,
-        web_search=0.0,
-        internal_reasoning=0.0,
-        max_prompt_cost=0.0,
-        max_completion_cost=0.0,
-        max_cost=0.0,
-    )
-
-    model = Model(
-        id="test-model-zero",
-        name="Test Model Zero",
-        created=1234567890,
-        description="A model with zero pricing",
-        context_length=8192,
-        architecture=base_architecture,
-        pricing=zero_pricing,
-    )
-
-    payload = _model_to_row_payload(model)
-    pricing_str = payload["pricing"]
-    assert isinstance(pricing_str, str)
-    pricing = json.loads(pricing_str)
-
-    assert pricing["prompt"] == pytest.approx(0.0, rel=1e-9)
-    assert pricing["completion"] == pytest.approx(0.0, rel=1e-9)
-    assert pricing["request"] == pytest.approx(0.0, rel=1e-9)
-
-
-def test_payload_structure_unchanged(standard_model: Model) -> None:
-    """Verify that payload structure matches expectations."""
-    payload = _model_to_row_payload(standard_model)
-
-    assert "id" in payload
-    assert "name" in payload
-    assert "created" in payload
-    assert "description" in payload
-    assert "context_length" in payload
-    assert "architecture" in payload
-    assert "pricing" in payload
-    assert "sats_pricing" in payload
-    assert "per_request_limits" in payload
-    assert "top_provider" in payload
-    assert "enabled" in payload
-    assert "upstream_provider_id" in payload
-
-    assert isinstance(payload["architecture"], str)
-    assert isinstance(payload["pricing"], str)
-
-
-def test_original_model_not_mutated(standard_model: Model) -> None:
-    """Verify that the original model object is not mutated."""
-    original_prompt = standard_model.pricing.prompt
-    original_completion = standard_model.pricing.completion
-
-    _model_to_row_payload(standard_model)
-
-    assert standard_model.pricing.prompt == original_prompt
-    assert standard_model.pricing.completion == original_completion

ui/Dockerfile

@@ -4,17 +4,22 @@ FROM base AS deps
 RUN apk add --no-cache libc6-compat
 WORKDIR /app
 
-COPY package.json yarn.lock* package-lock.json* pnpm-lock.yaml* .npmrc* ./
-RUN npm i
+RUN corepack enable pnpm && corepack prepare pnpm@latest --activate
+
+COPY package.json pnpm-lock.yaml* ./
+RUN pnpm install --frozen-lockfile
 
 FROM base AS builder
 WORKDIR /app
+
+RUN corepack enable pnpm && corepack prepare pnpm@latest --activate
+
 COPY --from=deps /app/node_modules ./node_modules
 COPY . .
 
 ENV NEXT_TELEMETRY_DISABLED=1
 
-RUN npm run build
+RUN pnpm run build
 FROM base AS runner
 WORKDIR /app
 

ui/Dockerfile.build

@@ -6,12 +6,14 @@ RUN apk add --no-cache libc6-compat
 WORKDIR /app
 
 # Copy package files
-COPY package.json package-lock.json* pnpm-lock.yaml* ./
-RUN npm ci
+COPY package.json pnpm-lock.yaml* ./
+RUN corepack enable pnpm && corepack prepare pnpm@latest --activate
+RUN pnpm install --frozen-lockfile
 
 # Build the UI
 FROM base AS builder
 WORKDIR /app
+RUN corepack enable pnpm && corepack prepare pnpm@latest --activate
 COPY --from=deps /app/node_modules ./node_modules
 COPY . .
 
@@ -27,7 +29,7 @@ ENV NODE_ENV=production
 ENV NEXT_TELEMETRY_DISABLED=1
 
 # Build the application
-RUN npm run build && \
+RUN pnpm run build && \
     echo "UI build completed at $(date)"
 
 # Use the builder stage as the final stage

ui/app/logs/log-filters.tsx

@@ -21,7 +21,17 @@ import {
   PopoverTrigger,
 } from '@/components/ui/popover';
 import { Calendar } from '@/components/ui/calendar';
-import { CalendarIcon, Filter, X } from 'lucide-react';
+import { Badge } from '@/components/ui/badge';
+import {
+  Command,
+  CommandEmpty,
+  CommandGroup,
+  CommandInput,
+  CommandItem,
+  CommandList,
+} from '@/components/ui/command';
+import { Checkbox } from '@/components/ui/checkbox';
+import { CalendarIcon, Filter, X, Plus } from 'lucide-react';
 import { useState, useEffect } from 'react';
 import { format } from 'date-fns';
 import { cn } from '@/lib/utils';
@@ -31,11 +41,17 @@ interface LogFiltersProps {
   selectedLevel: string;
   requestId: string;
   searchText: string;
+  selectedStatusCodes: string[];
+  selectedMethods: string[];
+  selectedEndpoints: string[];
   limit: number;
   onDateChange: (date: string) => void;
   onLevelChange: (level: string) => void;
   onRequestIdChange: (requestId: string) => void;
   onSearchTextChange: (searchText: string) => void;
+  onStatusCodesChange: (statusCodes: string[]) => void;
+  onMethodsChange: (methods: string[]) => void;
+  onEndpointsChange: (endpoints: string[]) => void;
   onLimitChange: (limit: number) => void;
   onClearFilters: () => void;
 }
@@ -43,16 +59,87 @@ interface LogFiltersProps {
 const LOG_LEVELS = ['TRACE', 'DEBUG', 'INFO', 'WARNING', 'ERROR', 'CRITICAL'];
 const PRESET_LIMITS = ['25', '50', '100', '200', '500', '1000'];
 
+const STATUS_CODE_OPTIONS = [
+  '200',
+  '201',
+  '204',
+  '400',
+  '401',
+  '402',
+  '403',
+  '404',
+  '422',
+  '429',
+  '500',
+  '502',
+  '503',
+  '504',
+];
+
+const METHOD_OPTIONS = [
+  'GET',
+  'POST',
+  'PUT',
+  'DELETE',
+  'PATCH',
+  'OPTIONS',
+  'HEAD',
+];
+
+const ENDPOINT_OPTIONS = [
+  '/chat/completions',
+  '/v1/chat/completions',
+  '/models',
+  '/v1/models',
+  '/responses',
+  '/v1/responses',
+  'v1/embeddings/models',
+  '/embeddings/models',
+];
+
+interface FilterBadgeProps {
+  value: string;
+  onRemove: (value: string) => void;
+}
+
+function FilterBadge({ value, onRemove }: FilterBadgeProps) {
+  return (
+    <Badge
+      variant='secondary'
+      className='flex items-center gap-1 px-1 font-normal'
+    >
+      {value}
+      <button
+        type='button'
+        onClick={(e) => {
+          e.preventDefault();
+          e.stopPropagation();
+          onRemove(value);
+        }}
+        className='hover:bg-muted-foreground/20 rounded-full'
+      >
+        <X className='h-3 w-3' />
+      </button>
+    </Badge>
+  );
+}
+
 export function LogFilters({
   selectedDate,
   selectedLevel,
   requestId,
   searchText,
+  selectedStatusCodes,
+  selectedMethods,
+  selectedEndpoints,
   limit,
   onDateChange,
   onLevelChange,
   onRequestIdChange,
   onSearchTextChange,
+  onStatusCodesChange,
+  onMethodsChange,
+  onEndpointsChange,
   onLimitChange,
   onClearFilters,
 }: LogFiltersProps) {
@@ -68,6 +155,10 @@ export function LogFilters({
       : undefined
   );
 
+  const [statusSearch, setStatusSearch] = useState('');
+  const [methodSearch, setMethodSearch] = useState('');
+  const [endpointSearch, setEndpointSearch] = useState('');
+
   useEffect(() => {
     const currentIsPreset = PRESET_LIMITS.includes(limit.toString());
     setIsCustom(!currentIsPreset);
@@ -129,6 +220,31 @@ export function LogFilters({
     }
   };
 
+  const toggleSelection = (
+    current: string[],
+    value: string,
+    onChange: (val: string[]) => void
+  ) => {
+    if (current.includes(value)) {
+      onChange(current.filter((v) => v !== value));
+    } else {
+      onChange([...current, value]);
+    }
+  };
+
+  const handleQuickStatusCode = (range: '4xx' | '5xx') => {
+    const codes = STATUS_CODE_OPTIONS.filter((c) => c.startsWith(range[0]));
+    const newSelection = new Set([...selectedStatusCodes]);
+    const allIncluded = codes.every((c) => selectedStatusCodes.includes(c));
+
+    if (allIncluded) {
+      codes.forEach((c) => newSelection.delete(c));
+    } else {
+      codes.forEach((c) => newSelection.add(c));
+    }
+    onStatusCodesChange(Array.from(newSelection));
+  };
+
   return (
     <Card className='mb-6'>
       <CardHeader>
@@ -137,7 +253,8 @@ export function LogFilters({
           Filters
         </CardTitle>
         <CardDescription>
-          Filter logs by date, level, request ID, text search, and limit
+          Filter logs by date, level, request ID, text search, status code,
+          method, endpoint and limit
         </CardDescription>
       </CardHeader>
       <CardContent>
@@ -197,6 +314,337 @@ export function LogFilters({
             </Select>
           </div>
 
+          <div className='space-y-2'>
+            <Label>Status Codes</Label>
+            <Popover>
+              <PopoverTrigger asChild>
+                <Button
+                  variant='outline'
+                  className='w-full justify-start text-left font-normal'
+                >
+                  <div className='flex flex-wrap gap-1'>
+                    {selectedStatusCodes.length > 0 ? (
+                      selectedStatusCodes.map((code) => (
+                        <FilterBadge
+                          key={code}
+                          value={code}
+                          onRemove={(val) =>
+                            toggleSelection(
+                              selectedStatusCodes,
+                              val,
+                              onStatusCodesChange
+                            )
+                          }
+                        />
+                      ))
+                    ) : (
+                      <span className='text-muted-foreground'>All codes</span>
+                    )}
+                  </div>
+                </Button>
+              </PopoverTrigger>
+              <PopoverContent className='w-64 p-0' align='start'>
+                <Command>
+                  <CommandInput
+                    placeholder='Search or add status code...'
+                    value={statusSearch}
+                    onValueChange={setStatusSearch}
+                  />
+                  <CommandList>
+                    {selectedStatusCodes.length > 0 && (
+                      <CommandGroup heading='Selected'>
+                        {selectedStatusCodes.map((code) => (
+                          <CommandItem
+                            key={`selected-${code}`}
+                            onSelect={() =>
+                              toggleSelection(
+                                selectedStatusCodes,
+                                code,
+                                onStatusCodesChange
+                              )
+                            }
+                          >
+                            <Checkbox checked={true} className='mr-2' />
+                            {code}
+                          </CommandItem>
+                        ))}
+                      </CommandGroup>
+                    )}
+                    {statusSearch &&
+                      !STATUS_CODE_OPTIONS.includes(statusSearch) &&
+                      !selectedStatusCodes.includes(statusSearch) && (
+                        <CommandGroup heading='Custom'>
+                          <CommandItem
+                            onSelect={() => {
+                              if (/^\d+$/.test(statusSearch)) {
+                                toggleSelection(
+                                  selectedStatusCodes,
+                                  statusSearch,
+                                  onStatusCodesChange
+                                );
+                                setStatusSearch('');
+                              }
+                            }}
+                          >
+                            <Plus className='mr-2 h-4 w-4' />
+                            Add &quot;{statusSearch}&quot;
+                          </CommandItem>
+                        </CommandGroup>
+                      )}
+                    <CommandEmpty>No results found.</CommandEmpty>
+                    <CommandGroup heading='Quick Filters'>
+                      <CommandItem
+                        onSelect={() => handleQuickStatusCode('4xx')}
+                      >
+                        <Checkbox
+                          checked={STATUS_CODE_OPTIONS.filter((c) =>
+                            c.startsWith('4')
+                          ).every((c) => selectedStatusCodes.includes(c))}
+                          className='mr-2'
+                        />
+                        4xx Errors
+                      </CommandItem>
+                      <CommandItem
+                        onSelect={() => handleQuickStatusCode('5xx')}
+                      >
+                        <Checkbox
+                          checked={STATUS_CODE_OPTIONS.filter((c) =>
+                            c.startsWith('5')
+                          ).every((c) => selectedStatusCodes.includes(c))}
+                          className='mr-2'
+                        />
+                        5xx Errors
+                      </CommandItem>
+                    </CommandGroup>
+                    <CommandGroup heading='Common Codes'>
+                      {STATUS_CODE_OPTIONS.filter(
+                        (code) => !selectedStatusCodes.includes(code)
+                      ).map((code) => (
+                        <CommandItem
+                          key={code}
+                          onSelect={() =>
+                            toggleSelection(
+                              selectedStatusCodes,
+                              code,
+                              onStatusCodesChange
+                            )
+                          }
+                        >
+                          <Checkbox checked={false} className='mr-2' />
+                          {code}
+                        </CommandItem>
+                      ))}
+                    </CommandGroup>
+                  </CommandList>
+                </Command>
+              </PopoverContent>
+            </Popover>
+          </div>
+
+          <div className='space-y-2'>
+            <Label>HTTP Methods</Label>
+            <Popover>
+              <PopoverTrigger asChild>
+                <Button
+                  variant='outline'
+                  className='w-full justify-start text-left font-normal'
+                >
+                  <div className='flex flex-wrap gap-1'>
+                    {selectedMethods.length > 0 ? (
+                      selectedMethods.map((method) => (
+                        <FilterBadge
+                          key={method}
+                          value={method}
+                          onRemove={(val) =>
+                            toggleSelection(
+                              selectedMethods,
+                              val,
+                              onMethodsChange
+                            )
+                          }
+                        />
+                      ))
+                    ) : (
+                      <span className='text-muted-foreground'>All methods</span>
+                    )}
+                  </div>
+                </Button>
+              </PopoverTrigger>
+              <PopoverContent className='w-64 p-0' align='start'>
+                <Command>
+                  <CommandInput
+                    placeholder='Search or add method...'
+                    value={methodSearch}
+                    onValueChange={setMethodSearch}
+                  />
+                  <CommandList>
+                    {selectedMethods.length > 0 && (
+                      <CommandGroup heading='Selected'>
+                        {selectedMethods.map((method) => (
+                          <CommandItem
+                            key={`selected-${method}`}
+                            onSelect={() =>
+                              toggleSelection(
+                                selectedMethods,
+                                method,
+                                onMethodsChange
+                              )
+                            }
+                          >
+                            <Checkbox checked={true} className='mr-2' />
+                            {method}
+                          </CommandItem>
+                        ))}
+                      </CommandGroup>
+                    )}
+                    {methodSearch &&
+                      !METHOD_OPTIONS.includes(methodSearch.toUpperCase()) &&
+                      !selectedMethods.includes(methodSearch.toUpperCase()) && (
+                        <CommandGroup heading='Custom'>
+                          <CommandItem
+                            onSelect={() => {
+                              toggleSelection(
+                                selectedMethods,
+                                methodSearch.toUpperCase(),
+                                onMethodsChange
+                              );
+                              setMethodSearch('');
+                            }}
+                          >
+                            <Plus className='mr-2 h-4 w-4' />
+                            Add &quot;{methodSearch.toUpperCase()}&quot;
+                          </CommandItem>
+                        </CommandGroup>
+                      )}
+                    <CommandEmpty>No results found.</CommandEmpty>
+                    <CommandGroup>
+                      {METHOD_OPTIONS.filter(
+                        (method) => !selectedMethods.includes(method)
+                      ).map((method) => (
+                        <CommandItem
+                          key={method}
+                          onSelect={() =>
+                            toggleSelection(
+                              selectedMethods,
+                              method,
+                              onMethodsChange
+                            )
+                          }
+                        >
+                          <Checkbox checked={false} className='mr-2' />
+                          {method}
+                        </CommandItem>
+                      ))}
+                    </CommandGroup>
+                  </CommandList>
+                </Command>
+              </PopoverContent>
+            </Popover>
+          </div>
+
+          <div className='space-y-2'>
+            <Label>Endpoints</Label>
+            <Popover>
+              <PopoverTrigger asChild>
+                <Button
+                  variant='outline'
+                  className='w-full justify-start text-left font-normal'
+                >
+                  <div className='flex flex-wrap gap-1 overflow-hidden'>
+                    {selectedEndpoints.length > 0 ? (
+                      selectedEndpoints.map((endpoint) => (
+                        <FilterBadge
+                          key={endpoint}
+                          value={endpoint}
+                          onRemove={(val) =>
+                            toggleSelection(
+                              selectedEndpoints,
+                              val,
+                              onEndpointsChange
+                            )
+                          }
+                        />
+                      ))
+                    ) : (
+                      <span className='text-muted-foreground'>
+                        All endpoints
+                      </span>
+                    )}
+                  </div>
+                </Button>
+              </PopoverTrigger>
+              <PopoverContent className='w-80 p-0' align='start'>
+                <Command>
+                  <CommandInput
+                    placeholder='Search or add endpoint pattern...'
+                    value={endpointSearch}
+                    onValueChange={setEndpointSearch}
+                  />
+                  <CommandList>
+                    {selectedEndpoints.length > 0 && (
+                      <CommandGroup heading='Selected'>
+                        {selectedEndpoints.map((endpoint) => (
+                          <CommandItem
+                            key={`selected-${endpoint}`}
+                            onSelect={() =>
+                              toggleSelection(
+                                selectedEndpoints,
+                                endpoint,
+                                onEndpointsChange
+                              )
+                            }
+                          >
+                            <Checkbox checked={true} className='mr-2' />
+                            {endpoint}
+                          </CommandItem>
+                        ))}
+                      </CommandGroup>
+                    )}
+                    {endpointSearch &&
+                      !ENDPOINT_OPTIONS.includes(endpointSearch) &&
+                      !selectedEndpoints.includes(endpointSearch) && (
+                        <CommandGroup heading='Custom'>
+                          <CommandItem
+                            onSelect={() => {
+                              toggleSelection(
+                                selectedEndpoints,
+                                endpointSearch,
+                                onEndpointsChange
+                              );
+                              setEndpointSearch('');
+                            }}
+                          >
+                            <Plus className='mr-2 h-4 w-4' />
+                            Add &quot;{endpointSearch}&quot;
+                          </CommandItem>
+                        </CommandGroup>
+                      )}
+                    <CommandEmpty>No results found.</CommandEmpty>
+                    <CommandGroup heading='Common Endpoints'>
+                      {ENDPOINT_OPTIONS.filter(
+                        (endpoint) => !selectedEndpoints.includes(endpoint)
+                      ).map((endpoint) => (
+                        <CommandItem
+                          key={endpoint}
+                          onSelect={() =>
+                            toggleSelection(
+                              selectedEndpoints,
+                              endpoint,
+                              onEndpointsChange
+                            )
+                          }
+                        >
+                          <Checkbox checked={false} className='mr-2' />
+                          {endpoint}
+                        </CommandItem>
+                      ))}
+                    </CommandGroup>
+                  </CommandList>
+                </Command>
+              </PopoverContent>
+            </Popover>
+          </div>
+
           <div className='space-y-2'>
             <Label htmlFor='request-id'>Request ID</Label>
             <Input

ui/app/logs/page.tsx

@@ -1,6 +1,6 @@
 'use client';
 
-import { useState } from 'react';
+import { useState, useEffect } from 'react';
 import { useQuery } from '@tanstack/react-query';
 import { AppSidebar } from '@/components/app-sidebar';
 import { SiteHeader } from '@/components/site-header';
@@ -22,15 +22,66 @@ import { LogFilters } from './log-filters';
 import { LogEntryCard } from './log-entry-card';
 import { LogDetailsDialog } from './log-details-dialog';
 
+const STORAGE_KEY = 'routstr-log-filters';
+
 export default function LogsPage() {
   const [selectedDate, setSelectedDate] = useState<string>('all');
   const [selectedLevel, setSelectedLevel] = useState<string>('all');
   const [requestId, setRequestId] = useState<string>('');
   const [searchText, setSearchText] = useState<string>('');
+  const [selectedStatusCodes, setSelectedStatusCodes] = useState<string[]>([]);
+  const [selectedMethods, setSelectedMethods] = useState<string[]>([]);
+  const [selectedEndpoints, setSelectedEndpoints] = useState<string[]>([]);
   const [limit, setLimit] = useState<number>(100);
   const [selectedLog, setSelectedLog] = useState<LogEntry | null>(null);
   const [isDialogOpen, setIsDialogOpen] = useState<boolean>(false);
 
+  // Load filters from localStorage on mount
+  useEffect(() => {
+    const saved = localStorage.getItem(STORAGE_KEY);
+    if (saved) {
+      try {
+        const parsed = JSON.parse(saved);
+        if (parsed.selectedDate) setSelectedDate(parsed.selectedDate);
+        if (parsed.selectedLevel) setSelectedLevel(parsed.selectedLevel);
+        if (parsed.requestId) setRequestId(parsed.requestId);
+        if (parsed.searchText) setSearchText(parsed.searchText);
+        if (parsed.selectedStatusCodes)
+          setSelectedStatusCodes(parsed.selectedStatusCodes);
+        if (parsed.selectedMethods) setSelectedMethods(parsed.selectedMethods);
+        if (parsed.selectedEndpoints)
+          setSelectedEndpoints(parsed.selectedEndpoints);
+        if (parsed.limit) setLimit(parsed.limit);
+      } catch (e) {
+        console.error('Failed to load filters from localStorage', e);
+      }
+    }
+  }, []);
+
+  // Save filters to localStorage whenever they change
+  useEffect(() => {
+    const filters = {
+      selectedDate,
+      selectedLevel,
+      requestId,
+      searchText,
+      selectedStatusCodes,
+      selectedMethods,
+      selectedEndpoints,
+      limit,
+    };
+    localStorage.setItem(STORAGE_KEY, JSON.stringify(filters));
+  }, [
+    selectedDate,
+    selectedLevel,
+    requestId,
+    searchText,
+    selectedStatusCodes,
+    selectedMethods,
+    selectedEndpoints,
+    limit,
+  ]);
+
   const {
     data: logsData,
     refetch: refetchLogs,
@@ -42,6 +93,9 @@ export default function LogsPage() {
       selectedLevel,
       requestId,
       searchText,
+      selectedStatusCodes,
+      selectedMethods,
+      selectedEndpoints,
       limit,
     ],
     queryFn: () =>
@@ -50,6 +104,16 @@ export default function LogsPage() {
         level: selectedLevel === 'all' ? undefined : selectedLevel,
         request_id: requestId || undefined,
         search: searchText || undefined,
+        status_codes:
+          selectedStatusCodes.length > 0
+            ? selectedStatusCodes.join(',')
+            : undefined,
+        methods:
+          selectedMethods.length > 0 ? selectedMethods.join(',') : undefined,
+        endpoints:
+          selectedEndpoints.length > 0
+            ? selectedEndpoints.join(',')
+            : undefined,
         limit: limit,
       }),
     refetchInterval: 30000,
@@ -60,6 +124,9 @@ export default function LogsPage() {
     setSelectedLevel('all');
     setRequestId('');
     setSearchText('');
+    setSelectedStatusCodes([]);
+    setSelectedMethods([]);
+    setSelectedEndpoints([]);
     setLimit(100);
   };
 
@@ -100,11 +167,17 @@ export default function LogsPage() {
             selectedLevel={selectedLevel}
             requestId={requestId}
             searchText={searchText}
+            selectedStatusCodes={selectedStatusCodes}
+            selectedMethods={selectedMethods}
+            selectedEndpoints={selectedEndpoints}
             limit={limit}
             onDateChange={setSelectedDate}
             onLevelChange={setSelectedLevel}
             onRequestIdChange={setRequestId}
             onSearchTextChange={setSearchText}
+            onStatusCodesChange={setSelectedStatusCodes}
+            onMethodsChange={setSelectedMethods}
+            onEndpointsChange={setSelectedEndpoints}
             onLimitChange={setLimit}
             onClearFilters={handleClearFilters}
           />
@@ -122,13 +195,22 @@ export default function LogsPage() {
               {(selectedDate !== 'all' ||
                 selectedLevel !== 'all' ||
                 requestId ||
-                searchText) && (
+                searchText ||
+                selectedStatusCodes.length > 0 ||
+                selectedMethods.length > 0 ||
+                selectedEndpoints.length > 0) && (
                 <CardDescription className='text-xs sm:text-sm'>
                   Showing logs
                   {selectedDate !== 'all' && ` for ${selectedDate}`}
                   {selectedLevel !== 'all' && ` with level ${selectedLevel}`}
                   {requestId && ` with request ID ${requestId}`}
                   {searchText && ` matching "${searchText}"`}
+                  {selectedStatusCodes.length > 0 &&
+                    ` with status ${selectedStatusCodes.join(', ')}`}
+                  {selectedMethods.length > 0 &&
+                    ` with method ${selectedMethods.join(', ')}`}
+                  {selectedEndpoints.length > 0 &&
+                    ` with endpoint ${selectedEndpoints.join(', ')}`}
                 </CardDescription>
               )}
             </CardHeader>

ui/app/logs/types.ts

@@ -17,6 +17,9 @@ export interface LogsResponse {
   level: string | null;
   request_id: string | null;
   search: string | null;
+  status_codes: string | null;
+  methods: string | null;
+  endpoints: string | null;
   limit: number;
 }
 

ui/app/providers/page.tsx

@@ -21,6 +21,7 @@ import {
   AdminModel,
 } from '@/lib/api/services/admin';
 import { AddProviderModelDialog } from '@/components/AddProviderModelDialog';
+import { BatchOverrideDialog } from '@/components/BatchOverrideDialog';
 import { Skeleton } from '@/components/ui/skeleton';
 import {
   AlertCircle,
@@ -208,7 +209,12 @@ function ProviderBalance({
     return <Skeleton className='h-9 w-24' />;
   }
 
-  if (error || !balanceData?.ok || !balanceData.balance_data) {
+  if (
+    error ||
+    !balanceData?.ok ||
+    balanceData.balance_data === undefined ||
+    balanceData.balance_data === null
+  ) {
     return null;
   }
 
@@ -281,7 +287,9 @@ function ProviderBalance({
               <div className='rounded-lg border-2 border-gray-200 p-2 dark:border-gray-800'>
                 {/* eslint-disable-next-line @next/next/no-img-element */}
                 <img
-                  src={`https://api.qrserver.com/v1/create-qr-code/?size=256x256&data=${encodeURIComponent(invoiceData.payment_request)}`}
+                  src={`https://api.qrserver.com/v1/create-qr-code/?size=256x256&data=${encodeURIComponent(
+                    invoiceData.payment_request
+                  )}`}
                   alt='Lightning Invoice QR Code'
                   className='h-64 w-64'
                 />
@@ -398,6 +406,9 @@ export default function ProvidersPage() {
     mode: 'create',
     initialData: null,
   });
+  const [batchOverrideProviderId, setBatchOverrideProviderId] = useState<
+    number | null
+  >(null);
 
   const [formData, setFormData] = useState<CreateUpstreamProvider>({
     provider_type: 'openrouter',
@@ -477,6 +488,25 @@ export default function ProvidersPage() {
     },
   });
 
+  const deleteModelMutation = useMutation({
+    mutationFn: ({
+      providerId,
+      modelId,
+    }: {
+      providerId: number;
+      modelId: string;
+    }) => AdminService.deleteProviderModel(providerId, modelId),
+    onSuccess: (_, variables) => {
+      queryClient.invalidateQueries({
+        queryKey: ['provider-models', variables.providerId],
+      });
+      toast.success('Model deleted successfully');
+    },
+    onError: (error: Error) => {
+      toast.error(`Failed to delete model: ${error.message}`);
+    },
+  });
+
   const handleCreateAccount = async () => {
     setIsCreatingAccount(true);
     try {
@@ -555,6 +585,12 @@ export default function ProvidersPage() {
     }
   };
 
+  const handleDeleteModel = (providerId: number, modelId: string) => {
+    if (confirm('Are you sure you want to delete this model?')) {
+      deleteModelMutation.mutate({ providerId, modelId });
+    }
+  };
+
   const getDefaultBaseUrl = (type: string) => {
     const providerType = providerTypes.find((pt) => pt.id === type);
     return providerType?.default_base_url || '';
@@ -622,6 +658,10 @@ export default function ProvidersPage() {
     });
   };
 
+  const handleBatchOverride = (providerId: number) => {
+    setBatchOverrideProviderId(providerId);
+  };
+
   return (
     <SidebarProvider>
       <AppSidebar variant='inset' />
@@ -928,15 +968,68 @@ export default function ProvidersPage() {
                               </div>
                             ) : providerModels &&
                               viewingModels === provider.id ? (
-                              providerModels.remote_models.length === 0 ? (
-                                // No provided models - show custom models directly without tabs
-                                <div className='space-y-2'>
-                                  {providerModels.db_models.length === 0 ? (
-                                    <div className='flex flex-col items-center justify-center gap-2 py-4'>
+                              <Tabs
+                                defaultValue={
+                                  providerModels.remote_models.length > 0
+                                    ? 'provided'
+                                    : 'custom'
+                                }
+                                className='w-full'
+                              >
+                                <TabsList className='grid w-full grid-cols-2'>
+                                  <TabsTrigger
+                                    value='provided'
+                                    className='text-xs sm:text-sm'
+                                  >
+                                    <span className='hidden sm:inline'>
+                                      Provided Models
+                                    </span>
+                                    <span className='sm:hidden'>Provided</span>
+                                    <Badge
+                                      variant='secondary'
+                                      className='ml-1 text-xs sm:ml-2'
+                                    >
+                                      {providerModels.remote_models.length}
+                                    </Badge>
+                                  </TabsTrigger>
+                                  <TabsTrigger
+                                    value='custom'
+                                    className='text-xs sm:text-sm'
+                                  >
+                                    <span className='hidden sm:inline'>
+                                      Custom Models
+                                    </span>
+                                    <span className='sm:hidden'>Custom</span>
+                                    <Badge
+                                      variant='secondary'
+                                      className='ml-1 text-xs sm:ml-2'
+                                    >
+                                      {providerModels.db_models.length}
+                                    </Badge>
+                                  </TabsTrigger>
+                                </TabsList>
+                                <TabsContent
+                                  value='custom'
+                                  className='mt-4 space-y-2'
+                                >
+                                  <div className='flex items-center justify-between'>
+                                    {providerModels.db_models.length > 0 && (
                                       <div className='text-muted-foreground text-sm'>
-                                        No models configured. Add custom models
-                                        to use this provider.
+                                        Custom models override or extend the
+                                        provider&apos;s catalog.
                                       </div>
+                                    )}
+                                    <div className='flex gap-2'>
+                                      <Button
+                                        variant='outline'
+                                        size='sm'
+                                        onClick={() =>
+                                          handleBatchOverride(provider.id)
+                                        }
+                                      >
+                                        <Database className='mr-2 h-4 w-4' />
+                                        Batch Override
+                                      </Button>
                                       <Button
                                         variant='outline'
                                         size='sm'
@@ -948,6 +1041,11 @@ export default function ProvidersPage() {
                                         Add Custom Model
                                       </Button>
                                     </div>
+                                  </div>
+                                  {providerModels.db_models.length === 0 ? (
+                                    <div className='text-muted-foreground py-4 text-center text-sm'>
+                                      No custom models configured
+                                    </div>
                                   ) : (
                                     <div className='space-y-2'>
                                       {providerModels.db_models.map((model) => (
@@ -995,103 +1093,48 @@ export default function ProvidersPage() {
                                             >
                                               <Pencil className='h-4 w-4' />
                                             </Button>
+                                            <Button
+                                              variant='ghost'
+                                              size='icon'
+                                              className='text-destructive hover:text-destructive h-8 w-8'
+                                              onClick={() =>
+                                                handleDeleteModel(
+                                                  provider.id,
+                                                  model.id
+                                                )
+                                              }
+                                              disabled={
+                                                deleteModelMutation.isPending
+                                              }
+                                            >
+                                              <Trash2 className='h-4 w-4' />
+                                            </Button>
                                           </div>
                                         </div>
                                       ))}
                                     </div>
                                   )}
-                                </div>
-                              ) : (
-                                // Has provided models - show tabs
-                                <Tabs
-                                  defaultValue='provided'
-                                  className='w-full'
+                                </TabsContent>
+                                <TabsContent
+                                  value='provided'
+                                  className='mt-4 space-y-2'
                                 >
-                                  <TabsList className='grid w-full grid-cols-2'>
-                                    <TabsTrigger
-                                      value='provided'
-                                      className='text-xs sm:text-sm'
-                                    >
-                                      <span className='hidden sm:inline'>
-                                        Provided Models
-                                      </span>
-                                      <span className='sm:hidden'>
-                                        Provided
-                                      </span>
-                                      <Badge
-                                        variant='secondary'
-                                        className='ml-1 text-xs sm:ml-2'
-                                      >
-                                        {providerModels.remote_models.length}
-                                      </Badge>
-                                    </TabsTrigger>
-                                    <TabsTrigger
-                                      value='custom'
-                                      className='text-xs sm:text-sm'
-                                    >
-                                      <span className='hidden sm:inline'>
-                                        Custom Models
-                                      </span>
-                                      <span className='sm:hidden'>Custom</span>
-                                      <Badge
-                                        variant='secondary'
-                                        className='ml-1 text-xs sm:ml-2'
-                                      >
-                                        {providerModels.db_models.length}
-                                      </Badge>
-                                    </TabsTrigger>
-                                  </TabsList>
-                                  <TabsContent
-                                    value='custom'
-                                    className='mt-4 space-y-2'
-                                  >
-                                    <div className='flex items-center justify-between'>
-                                      {providerModels.db_models.length > 0 && (
-                                        <div className='text-muted-foreground text-sm'>
-                                          Custom models override or extend the
-                                          provider&apos;s catalog.
-                                        </div>
-                                      )}
-                                      <Button
-                                        variant='outline'
-                                        size='sm'
-                                        onClick={() =>
-                                          handleAddModel(provider.id)
-                                        }
-                                      >
-                                        <Plus className='mr-2 h-4 w-4' />
-                                        Add
-                                      </Button>
-                                    </div>
-                                    {providerModels.db_models.length === 0 ? (
-                                      <div className='text-muted-foreground py-4 text-center text-sm'>
-                                        No custom models configured
+                                  {providerModels.remote_models.length > 0 ? (
+                                    <>
+                                      <div className='text-muted-foreground mb-3 text-sm'>
+                                        Models automatically discovered from the
+                                        provider&apos;s catalog.
                                       </div>
-                                    ) : (
                                       <div className='space-y-2'>
-                                        {providerModels.db_models.map(
+                                        {providerModels.remote_models.map(
                                           (model) => (
                                             <div
                                               key={model.id}
                                               className='hover:bg-accent flex flex-col gap-2 rounded-lg border p-3 transition-colors sm:flex-row sm:items-center sm:justify-between'
                                             >
                                               <div className='min-w-0 flex-1'>
-                                                <div className='flex flex-col gap-1 sm:flex-row sm:items-center sm:gap-2'>
-                                                  <span className='truncate font-mono text-sm font-medium'>
-                                                    {model.id}
-                                                  </span>
-                                                  <Badge
-                                                    variant={
-                                                      model.enabled
-                                                        ? 'default'
-                                                        : 'secondary'
-                                                    }
-                                                    className='w-fit text-xs'
-                                                  >
-                                                    {model.enabled
-                                                      ? 'Enabled'
-                                                      : 'Disabled'}
-                                                  </Badge>
+                                                <div className='truncate font-mono text-sm font-medium'>
+                                                  {model.id}
                                                 </div>
                                                 <div className='text-muted-foreground mt-1 text-xs break-words'>
                                                   {model.description ||
@@ -1104,79 +1147,32 @@ export default function ProvidersPage() {
                                                   tokens
                                                 </div>
                                                 <Button
-                                                  variant='ghost'
-                                                  size='icon'
-                                                  className='h-8 w-8'
+                                                  variant='outline'
+                                                  size='sm'
+                                                  className='h-7 text-xs'
                                                   onClick={() =>
-                                                    handleEditModel(
+                                                    handleOverrideModel(
                                                       provider.id,
                                                       model
                                                     )
                                                   }
                                                 >
-                                                  <Pencil className='h-4 w-4' />
+                                                  <Plus className='mr-1 h-3 w-3' />
+                                                  Override
                                                 </Button>
                                               </div>
                                             </div>
                                           )
                                         )}
                                       </div>
-                                    )}
-                                  </TabsContent>
-                                  <TabsContent
-                                    value='provided'
-                                    className='mt-4 space-y-2'
-                                  >
-                                    {providerModels.remote_models.length >
-                                      0 && (
-                                      <div className='text-muted-foreground mb-3 text-sm'>
-                                        Models automatically discovered from the
-                                        provider&apos;s catalog.
-                                      </div>
-                                    )}
-                                    <div className='space-y-2'>
-                                      {providerModels.remote_models.map(
-                                        (model) => (
-                                          <div
-                                            key={model.id}
-                                            className='hover:bg-accent flex flex-col gap-2 rounded-lg border p-3 transition-colors sm:flex-row sm:items-center sm:justify-between'
-                                          >
-                                            <div className='min-w-0 flex-1'>
-                                              <div className='truncate font-mono text-sm font-medium'>
-                                                {model.id}
-                                              </div>
-                                              <div className='text-muted-foreground mt-1 text-xs break-words'>
-                                                {model.description ||
-                                                  model.name}
-                                              </div>
-                                            </div>
-                                            <div className='flex items-center gap-2'>
-                                              <div className='text-muted-foreground text-xs whitespace-nowrap'>
-                                                {model.context_length?.toLocaleString()}{' '}
-                                                tokens
-                                              </div>
-                                              <Button
-                                                variant='outline'
-                                                size='sm'
-                                                className='h-7 text-xs'
-                                                onClick={() =>
-                                                  handleOverrideModel(
-                                                    provider.id,
-                                                    model
-                                                  )
-                                                }
-                                              >
-                                                <Plus className='mr-1 h-3 w-3' />
-                                                Override
-                                              </Button>
-                                            </div>
-                                          </div>
-                                        )
-                                      )}
+                                    </>
+                                  ) : (
+                                    <div className='text-muted-foreground py-4 text-center text-sm'>
+                                      No provided models available
                                     </div>
-                                  </TabsContent>
-                                </Tabs>
-                              )
+                                  )}
+                                </TabsContent>
+                              </Tabs>
                             ) : null}
                           </div>
                         )}
@@ -1352,6 +1348,19 @@ export default function ProvidersPage() {
             mode={modelDialogState.mode}
           />
         )}
+
+        {batchOverrideProviderId && (
+          <BatchOverrideDialog
+            providerId={batchOverrideProviderId}
+            isOpen={!!batchOverrideProviderId}
+            onClose={() => setBatchOverrideProviderId(null)}
+            onSuccess={() => {
+              queryClient.invalidateQueries({
+                queryKey: ['provider-models', batchOverrideProviderId],
+              });
+            }}
+          />
+        )}
       </SidebarInset>
     </SidebarProvider>
   );

ui/components/AddProviderModelDialog.tsx

@@ -248,7 +248,9 @@ export function AddProviderModelDialog({
     const pricing = model.pricing as Record<string, number>;
     const topProvider = model.top_provider as Record<string, unknown> | null;
 
-    form.setValue('id', model.id);
+    if (!isOverride) {
+      form.setValue('id', model.id);
+    }
     form.setValue('name', model.name);
     form.setValue('description', model.description || '');
     form.setValue('context_length', model.context_length);
@@ -425,7 +427,7 @@ export function AddProviderModelDialog({
           </DialogTitle>
           <DialogDescription>{description}</DialogDescription>
         </DialogHeader>
-        {!isEdit && !isOverride && (
+        {!isEdit && (
           <div className='bg-muted/30 rounded-md border p-3'>
             <div className='mb-2 text-sm font-medium'>Presets</div>
             <div className='grid gap-2 sm:grid-cols-3 sm:items-start'>
@@ -488,8 +490,9 @@ export function AddProviderModelDialog({
               </Popover>
             </div>
             <div className='text-muted-foreground mt-1 text-xs'>
-              Prefill fields from a preset model definition, then adjust as
-              needed.
+              {isOverride
+                ? 'Apply pricing and settings from a preset model (keeping the model ID unchanged).'
+                : 'Prefill fields from a preset model definition, then adjust as needed.'}
             </div>
           </div>
         )}
@@ -805,45 +808,6 @@ export function AddProviderModelDialog({
                     </FormItem>
                   )}
                 />
-                <FormField
-                  control={form.control}
-                  name='max_prompt_cost'
-                  render={({ field }) => (
-                    <FormItem>
-                      <FormLabel>Max Prompt Cost</FormLabel>
-                      <FormControl>
-                        <Input type='number' step='0.0001' {...field} />
-                      </FormControl>
-                      <FormMessage />
-                    </FormItem>
-                  )}
-                />
-                <FormField
-                  control={form.control}
-                  name='max_completion_cost'
-                  render={({ field }) => (
-                    <FormItem>
-                      <FormLabel>Max Completion Cost</FormLabel>
-                      <FormControl>
-                        <Input type='number' step='0.0001' {...field} />
-                      </FormControl>
-                      <FormMessage />
-                    </FormItem>
-                  )}
-                />
-                <FormField
-                  control={form.control}
-                  name='max_cost'
-                  render={({ field }) => (
-                    <FormItem>
-                      <FormLabel>Max Total Cost</FormLabel>
-                      <FormControl>
-                        <Input type='number' step='0.0001' {...field} />
-                      </FormControl>
-                      <FormMessage />
-                    </FormItem>
-                  )}
-                />
               </div>
             </div>
 

ui/components/BatchOverrideDialog.tsx

@@ -0,0 +1,144 @@
+'use client';
+
+import React, { useState } from 'react';
+import { Button } from '@/components/ui/button';
+import {
+  Dialog,
+  DialogContent,
+  DialogDescription,
+  DialogFooter,
+  DialogHeader,
+  DialogTitle,
+} from '@/components/ui/dialog';
+import { Textarea } from '@/components/ui/textarea';
+import { toast } from 'sonner';
+import { AdminService } from '@/lib/api/services/admin';
+import { Loader2, Database } from 'lucide-react';
+
+export interface BatchOverrideDialogProps {
+  providerId: number;
+  isOpen: boolean;
+  onClose: () => void;
+  onSuccess: () => void;
+}
+
+export function BatchOverrideDialog({
+  providerId,
+  isOpen,
+  onClose,
+  onSuccess,
+}: BatchOverrideDialogProps) {
+  const [jsonInput, setJsonInput] = useState('');
+  const [isSubmitting, setIsSubmitting] = useState(false);
+
+  const sampleJson = {
+    models: [
+      {
+        id: 'model-id-1',
+        name: 'Model Name 1',
+        description: 'Description...',
+        created: Math.floor(Date.now() / 1000),
+        context_length: 8192,
+        architecture: {
+          modality: 'text',
+          input_modalities: ['text'],
+          output_modalities: ['text'],
+          tokenizer: '',
+          instruct_type: null,
+        },
+        pricing: {
+          prompt: 0.0,
+          completion: 0.0,
+          request: 0.0,
+          image: 0.0,
+          web_search: 0.0,
+          internal_reasoning: 0.0,
+        },
+        enabled: true,
+      },
+    ],
+  };
+
+  const handleBatchOverride = async () => {
+    if (!jsonInput.trim()) {
+      toast.error('Please enter JSON content');
+      return;
+    }
+
+    setIsSubmitting(true);
+    try {
+      let data;
+      try {
+        data = JSON.parse(jsonInput);
+      } catch {
+        throw new Error('Invalid JSON format');
+      }
+
+      if (!data.models || !Array.isArray(data.models)) {
+        throw new Error('JSON match follow structure: { "models": [...] }');
+      }
+
+      const result = await AdminService.batchOverrideProviderModels(
+        providerId,
+        data.models
+      );
+
+      if (result.ok) {
+        toast.success(result.message || 'Batch override successful');
+        onSuccess();
+        onClose();
+        setJsonInput('');
+      } else {
+        throw new Error('Batch override failed');
+      }
+    } catch (error: unknown) {
+      const message =
+        error instanceof Error ? error.message : 'Batch override failed';
+      toast.error(message);
+    } finally {
+      setIsSubmitting(false);
+    }
+  };
+
+  return (
+    <Dialog open={isOpen} onOpenChange={(open) => !open && onClose()}>
+      <DialogContent className='sm:max-w-[800px]'>
+        <DialogHeader>
+          <DialogTitle className='flex items-center gap-2'>
+            <Database className='h-4 w-4' />
+            Batch Override Models
+          </DialogTitle>
+          <DialogDescription>
+            Paste a JSON object with a &quot;models&quot; array containing model
+            definitions. Existing models with the same ID will be updated.
+          </DialogDescription>
+        </DialogHeader>
+
+        <div className='grid gap-4 py-4'>
+          <Textarea
+            value={jsonInput}
+            onChange={(e) => setJsonInput(e.target.value)}
+            placeholder={JSON.stringify(sampleJson, null, 2)}
+            className='min-h-[400px] font-mono text-xs'
+          />
+        </div>
+
+        <DialogFooter>
+          <Button variant='outline' onClick={onClose}>
+            Cancel
+          </Button>
+          <Button onClick={handleBatchOverride} disabled={isSubmitting}>
+            {isSubmitting ? (
+              <>
+                <Loader2 className='mr-2 h-4 w-4 animate-spin' />
+                Processing...
+              </>
+            ) : (
+              'Batch Override'
+            )}
+          </Button>
+        </DialogFooter>
+      </DialogContent>
+    </Dialog>
+  );
+}

ui/components/child-key-creator.tsx

@@ -0,0 +1,546 @@
+'use client';
+
+import { useState } from 'react';
+import { WalletService } from '@/lib/api/services/wallet';
+import { Button } from '@/components/ui/button';
+import {
+  Card,
+  CardContent,
+  CardDescription,
+  CardHeader,
+  CardTitle,
+} from '@/components/ui/card';
+import { Alert, AlertDescription, AlertTitle } from '@/components/ui/alert';
+import { Input } from '@/components/ui/input';
+import {
+  Key,
+  Copy,
+  Check,
+  Loader2,
+  RotateCcw,
+  Plus,
+  Trash2,
+} from 'lucide-react';
+import { toast } from 'sonner';
+import { Badge } from '@/components/ui/badge';
+import { KeyOptions } from './key-options';
+
+interface KeyConfig {
+  id: string;
+  count: number;
+  balanceLimit: string;
+  balanceLimitReset: string;
+  validityDate: string;
+}
+
+interface ChildKeyCreatorProps {
+  baseUrl?: string;
+  apiKey?: string;
+  onApiKeyChange?: (apiKey: string) => void;
+  costPerKeyMsats?: number;
+}
+
+export function ChildKeyCreator({
+  baseUrl,
+  apiKey: propApiKey,
+  onApiKeyChange,
+  costPerKeyMsats,
+}: ChildKeyCreatorProps) {
+  const [internalApiKey, setInternalApiKey] = useState('');
+  const [loading, setLoading] = useState(false);
+  const [configs, setConfigs] = useState<KeyConfig[]>([
+    {
+      id: crypto.randomUUID(),
+      count: 1,
+      balanceLimit: '',
+      balanceLimitReset: '',
+      validityDate: '',
+    },
+  ]);
+  const [childKeyToCheck, setChildKeyToCheck] = useState('');
+  const [checking, setChecking] = useState(false);
+  const [keyStatus, setKeyStatus] = useState<{
+    total_spent: number;
+    balance_limit: number | null;
+    validity_date: number | null;
+    is_expired: boolean;
+    is_drained: boolean;
+  } | null>(null);
+  const [newKeys, setNewKeys] = useState<string[]>([]);
+  const [resultInfo, setResultInfo] = useState<{
+    cost_msats: number;
+    parent_balance: number;
+  } | null>(null);
+  const [copiedKey, setCopiedKey] = useState<string | null>(null);
+
+  const activeApiKey = propApiKey ?? internalApiKey;
+
+  const handleApiKeyChange = (val: string) => {
+    setInternalApiKey(val);
+    onApiKeyChange?.(val);
+  };
+
+  const addConfig = () => {
+    setConfigs([
+      ...configs,
+      {
+        id: crypto.randomUUID(),
+        count: 1,
+        balanceLimit: '',
+        balanceLimitReset: '',
+        validityDate: '',
+      },
+    ]);
+  };
+
+  const removeConfig = (id: string) => {
+    if (configs.length > 1) {
+      setConfigs(configs.filter((c) => c.id !== id));
+    }
+  };
+
+  const updateConfig = (id: string, updates: Partial<KeyConfig>) => {
+    setConfigs(configs.map((c) => (c.id === id ? { ...c, ...updates } : c)));
+  };
+
+  const handleCreateKey = async () => {
+    if (!activeApiKey && baseUrl) {
+      toast.error('Please provide a Parent API key first');
+      return;
+    }
+
+    setLoading(true);
+    try {
+      let allNewKeys: string[] = [];
+      let totalCost = 0;
+      let lastParentBalance = 0;
+
+      for (const config of configs) {
+        const requestedCount = Math.max(1, Math.min(50, Number(config.count)));
+        const result = await WalletService.createChildKey(
+          baseUrl,
+          activeApiKey,
+          requestedCount,
+          config.balanceLimit ? parseInt(config.balanceLimit) : undefined,
+          config.balanceLimitReset || undefined,
+          config.validityDate
+            ? Math.floor(
+                new Date(config.validityDate + 'T23:59:59').getTime() / 1000
+              )
+            : undefined
+        );
+
+        if (result.api_keys) {
+          allNewKeys = [...allNewKeys, ...result.api_keys];
+        }
+        totalCost += result.cost_msats;
+        lastParentBalance = result.parent_balance;
+      }
+
+      setNewKeys(allNewKeys);
+      setResultInfo({
+        cost_msats: totalCost,
+        parent_balance: lastParentBalance,
+      });
+
+      toast.success(
+        `${allNewKeys.length} child API key${
+          allNewKeys.length > 1 ? 's' : ''
+        } created successfully`
+      );
+    } catch (error) {
+      console.error('Failed to create child key:', error);
+      toast.error(
+        error instanceof Error ? error.message : 'Failed to create child key'
+      );
+    } finally {
+      setLoading(false);
+    }
+  };
+
+  const handleCheckKey = async () => {
+    if (!childKeyToCheck) {
+      toast.error('Please provide a Child API key to check');
+      return;
+    }
+
+    setChecking(true);
+    setKeyStatus(null);
+    try {
+      const baseUrlToUse = baseUrl || '';
+      const response = await fetch(`${baseUrlToUse}/v1/balance/info`, {
+        headers: {
+          Authorization: `Bearer ${childKeyToCheck}`,
+        },
+      });
+
+      if (!response.ok) {
+        throw new Error('Failed to fetch key info');
+      }
+
+      const info = await response.json();
+      const now = Math.floor(Date.now() / 1000);
+
+      setKeyStatus({
+        total_spent: info.total_spent,
+        balance_limit: info.balance_limit,
+        validity_date: info.validity_date,
+        is_expired: info.validity_date ? now > info.validity_date : false,
+        is_drained: info.balance_limit
+          ? info.total_spent >= info.balance_limit
+          : false,
+      });
+    } catch (error) {
+      toast.error(
+        error instanceof Error ? error.message : 'Failed to check child key'
+      );
+    } finally {
+      setChecking(false);
+    }
+  };
+
+  const copyToClipboard = (key: string) => {
+    navigator.clipboard.writeText(key);
+    setCopiedKey(key);
+    toast.success('API key copied to clipboard');
+    setTimeout(() => setCopiedKey(null), 2000);
+  };
+
+  const copyAllToClipboard = () => {
+    navigator.clipboard.writeText(newKeys.join('\n'));
+    toast.success('All API keys copied to clipboard');
+  };
+
+  return (
+    <div className='space-y-6'>
+      <Card>
+        <CardHeader>
+          <div className='flex items-center justify-between'>
+            <div className='space-y-1'>
+              <CardTitle>Create Child API Key</CardTitle>
+              <CardDescription>
+                Generate secondary API keys that share your account balance.
+              </CardDescription>
+            </div>
+            {costPerKeyMsats !== undefined && (
+              <div className='text-right'>
+                <p className='text-muted-foreground text-[0.65rem] tracking-wide uppercase'>
+                  Unit Cost
+                </p>
+                <p className='text-primary text-sm font-bold'>
+                  {costPerKeyMsats / 1000} sats
+                </p>
+              </div>
+            )}
+          </div>
+        </CardHeader>
+        <CardContent>
+          <div className='space-y-4'>
+            {baseUrl && (
+              <div className='space-y-2'>
+                <label className='text-muted-foreground text-[0.7rem] tracking-wider uppercase'>
+                  Parent API Key
+                </label>
+                <Input
+                  value={activeApiKey}
+                  onChange={(e) => handleApiKeyChange(e.target.value)}
+                  placeholder='sk-...'
+                  className='font-mono text-sm'
+                />
+              </div>
+            )}
+
+            <div className='flex flex-col gap-6'>
+              {configs.map((config, index) => (
+                <div
+                  key={config.id}
+                  className='bg-muted/30 relative space-y-4 rounded-lg border p-4 pt-6'
+                >
+                  {configs.length > 1 && (
+                    <Button
+                      variant='ghost'
+                      size='icon'
+                      className='text-destructive hover:bg-destructive/10 hover:text-destructive absolute top-2 right-2 h-7 w-7'
+                      onClick={() => removeConfig(config.id)}
+                    >
+                      <Trash2 className='h-4 w-4' />
+                    </Button>
+                  )}
+                  <div className='flex flex-col gap-4 sm:flex-row sm:items-end'>
+                    <div className='w-full space-y-2 sm:w-32'>
+                      <label className='text-muted-foreground text-[0.7rem] tracking-wider uppercase'>
+                        Number of keys
+                      </label>
+                      <Input
+                        type='number'
+                        min={1}
+                        max={50}
+                        value={config.count}
+                        onChange={(e) => {
+                          const val = parseInt(e.target.value);
+                          updateConfig(config.id, {
+                            count: isNaN(val)
+                              ? 1
+                              : Math.max(1, Math.min(50, val)),
+                          });
+                        }}
+                        className='h-9'
+                      />
+                    </div>
+
+                    <div className='flex-1'>
+                      <KeyOptions
+                        balanceLimit={config.balanceLimit}
+                        setBalanceLimit={(val) =>
+                          updateConfig(config.id, { balanceLimit: val })
+                        }
+                        validityDate={config.validityDate}
+                        setValidityDate={(val) =>
+                          updateConfig(config.id, { validityDate: val })
+                        }
+                        balanceLimitReset={config.balanceLimitReset}
+                        setBalanceLimitReset={(val) =>
+                          updateConfig(config.id, { balanceLimitReset: val })
+                        }
+                      />
+                    </div>
+                  </div>
+                </div>
+              ))}
+
+              <div className='flex justify-center'>
+                <Button
+                  variant='outline'
+                  size='sm'
+                  onClick={addConfig}
+                  className='gap-2 border-dashed'
+                >
+                  <Plus className='h-4 w-4' />
+                  Add Another Configuration
+                </Button>
+              </div>
+
+              <div className='flex flex-wrap items-center justify-between gap-4'>
+                <div className='text-muted-foreground text-xs'>
+                  {costPerKeyMsats && (
+                    <p>
+                      Total Cost:{' '}
+                      <span className='text-foreground font-medium'>
+                        {costPerKeyMsats *
+                          configs.reduce(
+                            (acc, c) => acc + Number(c.count),
+                            0
+                          )}{' '}
+                        mSats
+                      </span>
+                    </p>
+                  )}
+                </div>
+
+                <Button
+                  onClick={handleCreateKey}
+                  disabled={loading || (!!baseUrl && !activeApiKey)}
+                  className='w-full min-w-[140px] sm:w-auto'
+                >
+                  {loading ? (
+                    <>
+                      <Loader2 className='mr-2 h-4 w-4 animate-spin' />
+                      Creating...
+                    </>
+                  ) : (
+                    <>
+                      <Key className='mr-2 h-4 w-4' />
+                      Generate{' '}
+                      {configs.reduce(
+                        (acc, c) => acc + Number(c.count),
+                        0
+                      )}{' '}
+                      Keys
+                    </>
+                  )}
+                </Button>
+              </div>
+            </div>
+
+            <p className='text-muted-foreground text-xs'>
+              Each key creation has a small one-time fee.
+            </p>
+
+            {newKeys.length > 0 && (
+              <div className='mt-6 space-y-4'>
+                <Alert className='border-green-200 bg-green-50 dark:border-green-900/20 dark:bg-green-900/10'>
+                  <AlertTitle className='text-green-800 dark:text-green-400'>
+                    {newKeys.length} New API Key{newKeys.length > 1 ? 's' : ''}{' '}
+                    Generated
+                  </AlertTitle>
+                  <AlertDescription className='text-green-700 dark:text-green-500'>
+                    Copy {newKeys.length > 1 ? 'these keys' : 'this key'} now.
+                    You won&apos;t be able to see them again.
+                    {resultInfo && (
+                      <div className='mt-2 font-medium opacity-80'>
+                        Total Cost: {resultInfo.cost_msats / 1000} sats | New
+                        Balance: {resultInfo.parent_balance / 1000} sats
+                      </div>
+                    )}
+                  </AlertDescription>
+                </Alert>
+
+                <div className='space-y-2'>
+                  <div className='flex items-center justify-between'>
+                    <span className='text-muted-foreground text-xs font-medium uppercase'>
+                      Generated Keys ({newKeys.length})
+                    </span>
+                    {newKeys.length > 1 && (
+                      <Button
+                        variant='ghost'
+                        size='sm'
+                        className='h-7 text-[10px] uppercase'
+                        onClick={copyAllToClipboard}
+                      >
+                        <Copy className='mr-1 h-3 w-3' />
+                        Copy All
+                      </Button>
+                    )}
+                  </div>
+                  <div className='grid gap-2'>
+                    {newKeys.map((key, index) => (
+                      <div
+                        key={index}
+                        className='group relative flex items-center gap-2'
+                      >
+                        <code className='bg-muted/50 flex-1 rounded border p-2.5 font-mono text-[10px] break-all sm:text-xs'>
+                          {key}
+                        </code>
+                        <Button
+                          size='icon'
+                          variant='ghost'
+                          className='h-8 w-8 shrink-0'
+                          onClick={() => copyToClipboard(key)}
+                        >
+                          {copiedKey === key ? (
+                            <Check className='h-3.5 w-3.5 text-green-500' />
+                          ) : (
+                            <Copy className='h-3.5 w-3.5 opacity-50 group-hover:opacity-100' />
+                          )}
+                        </Button>
+                      </div>
+                    ))}
+                  </div>
+                </div>
+
+                {newKeys.length > 3 && (
+                  <div className='space-y-2'>
+                    <label className='text-muted-foreground text-[0.7rem] tracking-wider uppercase'>
+                      Bulk Export (All Keys)
+                    </label>
+                    <div className='relative'>
+                      <textarea
+                        readOnly
+                        value={newKeys.join('\n')}
+                        rows={Math.min(newKeys.length, 6)}
+                        className='bg-muted/30 w-full rounded-md border p-3 font-mono text-[10px] focus:outline-none'
+                      />
+                      <Button
+                        size='sm'
+                        variant='secondary'
+                        className='absolute right-2 bottom-2 h-7 text-[10px]'
+                        onClick={copyAllToClipboard}
+                      >
+                        Copy Bulk
+                      </Button>
+                    </div>
+                  </div>
+                )}
+              </div>
+            )}
+          </div>
+        </CardContent>
+      </Card>
+
+      <Card>
+        <CardHeader>
+          <CardTitle className='text-lg'>Check Child Key Status</CardTitle>
+          <CardDescription>
+            View the current spending, limit, and expiration status of any child
+            key.
+          </CardDescription>
+        </CardHeader>
+        <CardContent>
+          <div className='space-y-4'>
+            <div className='space-y-2'>
+              <label className='text-muted-foreground text-[0.7rem] tracking-wider uppercase'>
+                Child API Key
+              </label>
+              <Input
+                value={childKeyToCheck}
+                onChange={(e) => setChildKeyToCheck(e.target.value)}
+                placeholder='sk-...'
+                className='font-mono text-sm'
+              />
+            </div>
+            <Button
+              onClick={handleCheckKey}
+              disabled={checking || !childKeyToCheck}
+              variant='outline'
+              className='w-full'
+            >
+              {checking ? (
+                <>
+                  <Loader2 className='mr-2 h-4 w-4 animate-spin' />
+                  Checking...
+                </>
+              ) : (
+                <>
+                  <RotateCcw className='mr-2 h-4 w-4' />
+                  Check Status
+                </>
+              )}
+            </Button>
+
+            {keyStatus && (
+              <div className='bg-muted/30 mt-4 space-y-3 rounded-lg border p-4 text-sm'>
+                <div className='flex justify-between'>
+                  <span className='text-muted-foreground'>Total Spent:</span>
+                  <span className='font-mono font-medium'>
+                    {keyStatus.total_spent} mSats
+                  </span>
+                </div>
+                {keyStatus.balance_limit !== null && (
+                  <div className='flex justify-between'>
+                    <span className='text-muted-foreground'>Limit:</span>
+                    <span className='font-mono font-medium'>
+                      {keyStatus.balance_limit} mSats
+                    </span>
+                  </div>
+                )}
+                {keyStatus.validity_date !== null && (
+                  <div className='flex justify-between'>
+                    <span className='text-muted-foreground'>Expires:</span>
+                    <span className='font-mono font-medium'>
+                      {new Date(
+                        keyStatus.validity_date * 1000
+                      ).toLocaleDateString()}
+                    </span>
+                  </div>
+                )}
+                <div className='flex gap-2 pt-2'>
+                  {keyStatus.is_drained && (
+                    <Badge variant='destructive'>Drained</Badge>
+                  )}
+                  {keyStatus.is_expired && (
+                    <Badge variant='destructive'>Expired</Badge>
+                  )}
+                  {!keyStatus.is_drained && !keyStatus.is_expired && (
+                    <Badge className='bg-green-600 hover:bg-green-700'>
+                      Active
+                    </Badge>
+                  )}
+                </div>
+              </div>
+            )}
+          </div>
+        </CardContent>
+      </Card>
+    </div>
+  );
+}

ui/components/key-options.tsx

@@ -0,0 +1,72 @@
+import { Zap, Calendar, Shield } from 'lucide-react';
+import { Input } from '@/components/ui/input';
+
+interface KeyOptionsProps {
+  balanceLimit: string;
+  setBalanceLimit: (val: string) => void;
+  validityDate: string;
+  setValidityDate: (val: string) => void;
+  balanceLimitReset: string;
+  setBalanceLimitReset: (val: string) => void;
+  showBalanceLimit?: boolean;
+}
+
+export function KeyOptions({
+  balanceLimit,
+  setBalanceLimit,
+  validityDate,
+  setValidityDate,
+  balanceLimitReset,
+  setBalanceLimitReset,
+  showBalanceLimit = true,
+}: KeyOptionsProps) {
+  return (
+    <div className='grid gap-4 sm:grid-cols-3'>
+      {showBalanceLimit && (
+        <div className='space-y-2'>
+          <label className='text-muted-foreground flex items-center gap-1.5 text-[0.7rem] tracking-wider uppercase'>
+            <Zap className='h-3 w-3' />
+            Balance Limit (mSats)
+          </label>
+          <Input
+            type='number'
+            placeholder='No limit'
+            value={balanceLimit}
+            onChange={(e) => setBalanceLimit(e.target.value)}
+            className='h-9 text-xs'
+          />
+        </div>
+      )}
+
+      <div className='space-y-2'>
+        <label className='text-muted-foreground flex items-center gap-1.5 text-[0.7rem] tracking-wider uppercase'>
+          <Calendar className='h-3 w-3' />
+          Validity Date
+        </label>
+        <Input
+          type='date'
+          value={validityDate}
+          onChange={(e) => setValidityDate(e.target.value)}
+          className='h-9 text-xs'
+        />
+      </div>
+
+      <div className='space-y-2'>
+        <label className='text-muted-foreground flex items-center gap-1.5 text-[0.7rem] tracking-wider uppercase'>
+          <Shield className='h-3 w-3' />
+          Reset Policy
+        </label>
+        <select
+          value={balanceLimitReset}
+          onChange={(e) => setBalanceLimitReset(e.target.value)}
+          className='bg-background border-input flex h-9 w-full rounded-md border px-3 py-1 text-xs shadow-sm transition-colors'
+        >
+          <option value=''>None</option>
+          <option value='daily'>Daily</option>
+          <option value='weekly'>Weekly</option>
+          <option value='monthly'>Monthly</option>
+        </select>
+      </div>
+    </div>
+  );
+}

ui/components/landing/cashu-payment-workflow.tsx

@@ -8,6 +8,7 @@ import { Button } from '@/components/ui/button';
 import { Textarea } from '@/components/ui/textarea';
 import { Input } from '@/components/ui/input';
 import { Separator } from '@/components/ui/separator';
+import { KeyOptions } from '@/components/key-options';
 
 type WalletSnapshot = {
   apiKey: string;
@@ -86,9 +87,11 @@ export function CashuPaymentWorkflow({
   const [isTopupLoading, setIsTopupLoading] = useState(false);
   const [isRefunding, setIsRefunding] = useState(false);
   const [isSyncingBalance, setIsSyncingBalance] = useState(false);
-  const [hasInteractedCreate, setHasInteractedCreate] = useState(false);
   const [hasInteractedManage, setHasInteractedManage] = useState(false);
   const [hasInteractedTopup, setHasInteractedTopup] = useState(false);
+  const [balanceLimit, setBalanceLimit] = useState<string>('');
+  const [balanceLimitReset, setBalanceLimitReset] = useState<string>('');
+  const [validityDate, setValidityDate] = useState<string>('');
 
   const activeApiKey = apiKeyInput.trim();
 
@@ -121,6 +124,15 @@ export function CashuPaymentWorkflow({
       const params = new URLSearchParams({
         initial_balance_token: initialToken.trim(),
       });
+      if (balanceLimit) params.append('balance_limit', balanceLimit);
+      if (balanceLimitReset)
+        params.append('balance_limit_reset', balanceLimitReset);
+      if (validityDate) {
+        const timestamp = Math.floor(
+          new Date(validityDate + 'T23:59:59').getTime() / 1000
+        );
+        params.append('validity_date', timestamp.toString());
+      }
       const response = await fetch(
         `${baseUrl}/v1/balance/create?${params.toString()}`,
         {
@@ -154,7 +166,14 @@ export function CashuPaymentWorkflow({
     } finally {
       setIsCreatingKey(false);
     }
-  }, [initialToken, baseUrl, onApiKeyCreated]);
+  }, [
+    initialToken,
+    baseUrl,
+    onApiKeyCreated,
+    balanceLimit,
+    balanceLimitReset,
+    validityDate,
+  ]);
 
   const handleSyncBalance = useCallback(async (): Promise<void> => {
     if (!activeApiKey) {
@@ -256,11 +275,10 @@ export function CashuPaymentWorkflow({
     [apiKey, onApiKeyChanged, onWalletInfoUpdated]
   );
 
-  const showCreateDetails =
-    hasInteractedCreate || initialToken.trim().length > 0;
   const showManageDetails = hasInteractedManage || Boolean(walletInfo);
   const showTopupDetails = hasInteractedTopup || topupToken.trim().length > 0;
   const canTopup = Boolean(activeApiKey);
+  const showCreateDetails = initialToken.trim().length > 0;
 
   return (
     <Card>
@@ -285,12 +303,21 @@ export function CashuPaymentWorkflow({
             value={initialToken}
             onChange={(event) => setInitialToken(event.target.value)}
             placeholder='cashuA1...'
-            rows={showCreateDetails ? 4 : 2}
+            rows={4}
             className='font-mono text-sm transition-all duration-200'
-            onFocus={() => setHasInteractedCreate(true)}
           />
-          {showCreateDetails && (
-            <div className='flex flex-wrap gap-2'>
+          <div className='space-y-4'>
+            <KeyOptions
+              balanceLimit={balanceLimit}
+              setBalanceLimit={setBalanceLimit}
+              validityDate={validityDate}
+              setValidityDate={setValidityDate}
+              balanceLimitReset={balanceLimitReset}
+              setBalanceLimitReset={setBalanceLimitReset}
+              showBalanceLimit={false}
+            />
+
+            <div className='flex flex-wrap items-center gap-3'>
               <Button
                 onClick={handleCreateKey}
                 disabled={isCreatingKey}
@@ -298,11 +325,13 @@ export function CashuPaymentWorkflow({
               >
                 {isCreatingKey ? 'Creating…' : 'Create API key'}
               </Button>
-              <span className='text-muted-foreground text-xs'>
+              <span className='text-muted-foreground text-[0.7rem] leading-relaxed'>
                 Redeems instantly and returns <code>sk-</code> key.
+                <br />
+                Optional limits can be set above for enhanced security.
               </span>
             </div>
-          )}
+          </div>
         </section>
 
         <Separator />

ui/components/landing/cheat-sheet.tsx

@@ -14,6 +14,7 @@ import { ConfigurationService } from '@/lib/api/services/configuration';
 import { CashuPaymentWorkflow } from './cashu-payment-workflow';
 import { LightningPaymentWorkflow } from './lightning-payment-workflow';
 import { ApiKeyManager } from './api-key-manager';
+import { ChildKeyCreator } from '@/components/child-key-creator';
 
 type NodeInfo = {
   name: string;
@@ -23,6 +24,7 @@ type NodeInfo = {
   mints: string[];
   http_url?: string | null;
   onion_url?: string | null;
+  child_key_cost_msats?: number;
 };
 
 type WalletSnapshot = {
@@ -362,10 +364,11 @@ export function CheatSheet(): JSX.Element {
         </section>
 
         <Tabs defaultValue='cashu' className='w-full'>
-          <TabsList className='grid w-full grid-cols-3'>
+          <TabsList className='grid w-full grid-cols-4'>
             <TabsTrigger value='cashu'>Cashu Payments</TabsTrigger>
             <TabsTrigger value='lightning'>Lightning Payments</TabsTrigger>
             <TabsTrigger value='manage'>Manage Keys</TabsTrigger>
+            <TabsTrigger value='child-keys'>Child Keys</TabsTrigger>
           </TabsList>
 
           <TabsContent value='cashu' className='space-y-4'>
@@ -422,6 +425,15 @@ export function CheatSheet(): JSX.Element {
               </Card>
             )}
           </TabsContent>
+
+          <TabsContent value='child-keys' className='space-y-4'>
+            <ChildKeyCreator
+              baseUrl={normalizedBaseUrl}
+              apiKey={apiKeyInput}
+              onApiKeyChange={handleApiKeyChanged}
+              costPerKeyMsats={nodeInfo?.child_key_cost_msats}
+            />
+          </TabsContent>
         </Tabs>
       </main>
     </div>

ui/components/landing/lightning-payment-workflow.tsx

@@ -10,6 +10,7 @@ import { Button } from '@/components/ui/button';
 import { Input } from '@/components/ui/input';
 import { Textarea } from '@/components/ui/textarea';
 import { Separator } from '@/components/ui/separator';
+import { KeyOptions } from '@/components/key-options';
 
 type WalletSnapshot = {
   apiKey: string;
@@ -89,7 +90,10 @@ export function LightningPaymentWorkflow({
   const [isTopupping, setIsTopupping] = useState(false);
   const [isRecovering, setIsRecovering] = useState(false);
 
-  const [hasInteractedCreate, setHasInteractedCreate] = useState(false);
+  const [balanceLimit, setBalanceLimit] = useState<string>('');
+  const [balanceLimitReset, setBalanceLimitReset] = useState<string>('');
+  const [validityDate, setValidityDate] = useState<string>('');
+
   const [hasInteractedTopup, setHasInteractedTopup] = useState(false);
   const [hasInteractedRecover, setHasInteractedRecover] = useState(false);
 
@@ -166,13 +170,29 @@ export function LightningPaymentWorkflow({
     setIsCreating(true);
 
     try {
+      const payload: {
+        amount_sats: number;
+        purpose: string;
+        balance_limit?: number;
+        balance_limit_reset?: string;
+        validity_date?: number;
+      } = {
+        amount_sats: amount,
+        purpose: 'create',
+      };
+
+      if (balanceLimit) payload.balance_limit = parseInt(balanceLimit);
+      if (balanceLimitReset) payload.balance_limit_reset = balanceLimitReset;
+      if (validityDate) {
+        payload.validity_date = Math.floor(
+          new Date(validityDate + 'T23:59:59').getTime() / 1000
+        );
+      }
+
       const response = await fetch(`${baseUrl}/v1/balance/lightning/invoice`, {
         method: 'POST',
         headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({
-          amount_sats: amount,
-          purpose: 'create',
-        }),
+        body: JSON.stringify(payload),
       });
 
       if (!response.ok) {
@@ -214,7 +234,15 @@ export function LightningPaymentWorkflow({
     } finally {
       setIsCreating(false);
     }
-  }, [createAmount, baseUrl, pollInvoiceStatus, onApiKeyCreated]);
+  }, [
+    createAmount,
+    baseUrl,
+    pollInvoiceStatus,
+    onApiKeyCreated,
+    balanceLimit,
+    balanceLimitReset,
+    validityDate,
+  ]);
 
   const handleTopupInvoice = useCallback(async (): Promise<void> => {
     const amount = parseInt(topupAmount);
@@ -333,14 +361,13 @@ export function LightningPaymentWorkflow({
     }
   }, [recoverInvoice, baseUrl, onApiKeyCreated]);
 
-  const showCreateDetails =
-    hasInteractedCreate || createAmount.trim().length > 0;
   const showTopupDetails =
     hasInteractedTopup ||
     topupAmount.trim().length > 0 ||
     topupApiKey.trim().length > 0;
   const showRecoverDetails =
     hasInteractedRecover || recoverInvoice.trim().length > 0;
+  const showCreateDetails = createAmount.trim().length > 0;
 
   return (
     <Card>
@@ -367,9 +394,18 @@ export function LightningPaymentWorkflow({
             onChange={(event) => setCreateAmount(event.target.value)}
             placeholder='Amount in sats (e.g., 1000)'
             className='text-sm'
-            onFocus={() => setHasInteractedCreate(true)}
           />
-          {showCreateDetails && (
+          <div className='space-y-4'>
+            <KeyOptions
+              balanceLimit={balanceLimit}
+              setBalanceLimit={setBalanceLimit}
+              validityDate={validityDate}
+              setValidityDate={setValidityDate}
+              balanceLimitReset={balanceLimitReset}
+              setBalanceLimitReset={setBalanceLimitReset}
+              showBalanceLimit={false}
+            />
+
             <div className='space-y-3'>
               <Button
                 onClick={handleCreateInvoice}
@@ -473,7 +509,7 @@ export function LightningPaymentWorkflow({
                 </div>
               )}
             </div>
-          )}
+          </div>
         </section>
 
         <Separator />

ui/components/temporary-balances.tsx

@@ -60,7 +60,11 @@ export function TemporaryBalances({
     let totalRequests = 0;
 
     balances.forEach((balance) => {
-      totalBalance += balance.balance || 0;
+      // Only count parents for total balance to avoid double counting
+      // since child keys use parent balance
+      if (!balance.parent_key_hash) {
+        totalBalance += balance.balance || 0;
+      }
       totalSpent += balance.total_spent || 0;
       totalRequests += balance.total_requests || 0;
     });
@@ -72,6 +76,34 @@ export function TemporaryBalances({
     ? calculateTotals(data)
     : { totalBalance: 0, totalSpent: 0, totalRequests: 0 };
 
+  // Group parents and children
+  const hierarchicalData = (() => {
+    if (!data) return [];
+
+    const parents = filteredData.filter((item) => !item.parent_key_hash);
+    const result: (TemporaryBalance & { isChild?: boolean })[] = [];
+
+    parents.forEach((parent) => {
+      result.push(parent);
+      const children = data.filter(
+        (item) => item.parent_key_hash === parent.hashed_key
+      );
+      children.forEach((child) => {
+        result.push({ ...child, isChild: true });
+      });
+    });
+
+    // Add children whose parents didn't match the search or aren't in the list
+    const orphans = filteredData.filter(
+      (item) =>
+        item.parent_key_hash &&
+        !result.some((r) => r.hashed_key === item.hashed_key)
+    );
+    result.push(...orphans.map((o) => ({ ...o, isChild: true })));
+
+    return result;
+  })();
+
   return (
     <>
       <Card className='h-full w-full shadow-sm'>
@@ -182,22 +214,37 @@ export function TemporaryBalances({
                   <div className='text-right'>Expiry Time</div>
                 </div>
 
-                {filteredData.length > 0 ? (
-                  filteredData.map((balance, index) => (
+                {hierarchicalData.length > 0 ? (
+                  hierarchicalData.map((balance, index) => (
                     <div
                       key={index}
                       className={cn(
                         'hover:bg-muted/50 border-t p-3 text-sm transition-colors',
-                        balance.balance === 0 && 'opacity-60'
+                        balance.balance === 0 &&
+                          !balance.isChild &&
+                          'opacity-60',
+                        balance.isChild &&
+                          'ml-4 border-l-2 border-l-blue-200 bg-blue-50/30'
                       )}
                     >
                       {/* Desktop Layout */}
                       <div className='hidden grid-cols-6 gap-2 md:grid'>
-                        <div className='max-w-32 truncate font-mono text-xs break-all'>
+                        <div className='flex max-w-48 items-center gap-2 truncate font-mono text-xs break-all'>
+                          {balance.isChild && (
+                            <span className='rounded bg-blue-100 px-1 py-0.5 text-[10px] font-bold text-blue-700 uppercase'>
+                              Child
+                            </span>
+                          )}
                           {balance.hashed_key}
                         </div>
                         <div className='text-right font-mono'>
-                          {formatBalance(balance.balance)}
+                          {balance.isChild ? (
+                            <span className='text-muted-foreground italic'>
+                              (Parent)
+                            </span>
+                          ) : (
+                            formatBalance(balance.balance)
+                          )}
                         </div>
                         <div className='text-right font-mono'>
                           {formatBalance(balance.total_spent)}
@@ -226,13 +273,20 @@ export function TemporaryBalances({
 
                       {/* Mobile Layout */}
                       <div className='space-y-3 md:hidden'>
-                        <div className='space-y-1'>
-                          <span className='text-muted-foreground text-xs font-medium'>
-                            Key
-                          </span>
-                          <div className='font-mono text-xs break-all'>
-                            {balance.hashed_key}
+                        <div className='flex items-center justify-between'>
+                          <div className='space-y-1'>
+                            <span className='text-muted-foreground text-xs font-medium'>
+                              {balance.isChild ? 'Child Key' : 'Key'}
+                            </span>
+                            <div className='font-mono text-xs break-all'>
+                              {balance.hashed_key}
+                            </div>
                           </div>
+                          {balance.isChild && (
+                            <span className='rounded bg-blue-100 px-1.5 py-0.5 text-[10px] font-bold text-blue-700 uppercase'>
+                              Child
+                            </span>
+                          )}
                         </div>
 
                         <div className='grid grid-cols-2 gap-3'>
@@ -241,7 +295,13 @@ export function TemporaryBalances({
                               Balance
                             </div>
                             <div className='truncate font-mono text-sm'>
-                              {formatBalance(balance.balance)}
+                              {balance.isChild ? (
+                                <span className='text-muted-foreground text-xs italic'>
+                                  (Uses Parent)
+                                </span>
+                              ) : (
+                                formatBalance(balance.balance)
+                              )}
                             </div>
                           </div>
                           <div className='space-y-1'>

ui/lib/api/services/admin.ts

@@ -139,18 +139,26 @@ export class AdminService {
   ): Record<string, unknown> {
     if (!pricing) return pricing;
     const result = { ...pricing };
-    if (typeof result.prompt === 'number') {
-      result.prompt = result.prompt * 1000000;
-    }
-    if (typeof result.completion === 'number') {
-      result.completion = result.completion * 1000000;
-    }
-    if (typeof result.request === 'number') {
-      result.request = result.request * 1000000;
-    }
-    if (typeof result.image === 'number') {
-      result.image = result.image * 1000000;
-    }
+
+    // Only prompt and completion are per-token and need scaling to per-1M
+    const convertField = (field: string) => {
+      const val = result[field];
+      if (val !== undefined && val !== null) {
+        const num = typeof val === 'string' ? parseFloat(val) : (val as number);
+        if (!isNaN(num)) {
+          // Multiply by 1M and round to avoid floating point artifacts (e.g. 0.40399999999999997)
+          // 9 decimals is plenty for USD/1M tokens (0.000000001)
+          result[field] = parseFloat((num * 1000000).toFixed(9));
+        }
+      }
+    };
+
+    convertField('prompt');
+    convertField('completion');
+
+    // Other fields (request, image, etc.) are already flat fees (per item)
+    // so we do NOT scale them.
+
     return result;
   }
 
@@ -159,18 +167,23 @@ export class AdminService {
   ): Record<string, unknown> {
     if (!pricing) return pricing;
     const result = { ...pricing };
-    if (typeof result.prompt === 'number') {
-      result.prompt = result.prompt / 1000000;
-    }
-    if (typeof result.completion === 'number') {
-      result.completion = result.completion / 1000000;
-    }
-    if (typeof result.request === 'number') {
-      result.request = result.request / 1000000;
-    }
-    if (typeof result.image === 'number') {
-      result.image = result.image / 1000000;
-    }
+
+    // Only prompt and completion are per-1M in UI and need scaling down to per-token
+    const convertField = (field: string) => {
+      const val = result[field];
+      if (val !== undefined && val !== null) {
+        const num = typeof val === 'string' ? parseFloat(val) : (val as number);
+        if (!isNaN(num)) {
+          result[field] = num / 1000000;
+        }
+      }
+    };
+
+    convertField('prompt');
+    convertField('completion');
+
+    // Other fields stay as flat fees
+
     return result;
   }
 
@@ -291,7 +304,19 @@ export class AdminService {
     const data = await apiClient.get<ProviderModels>(
       `/admin/api/upstream-providers/${providerId}/models`
     );
-    return data;
+
+    // Convert pricing for all models in the list so the UI receives "per 1M tokens" values
+    return {
+      ...data,
+      db_models: data.db_models.map((m) => ({
+        ...m,
+        pricing: this.convertPricingToPerMillionTokens(m.pricing),
+      })),
+      remote_models: data.remote_models.map((m) => ({
+        ...m,
+        pricing: this.convertPricingToPerMillionTokens(m.pricing),
+      })),
+    };
   }
 
   static async createProviderModel(
@@ -316,6 +341,23 @@ export class AdminService {
     };
   }
 
+  static async batchOverrideProviderModels(
+    providerId: number,
+    models: AdminModel[]
+  ): Promise<{ ok: boolean; count: number; message: string }> {
+    const payload = {
+      models: models.map((m) => ({
+        ...m,
+        pricing: this.convertPricingToPerToken(m.pricing),
+      })),
+    };
+    return await apiClient.post<{
+      ok: boolean;
+      count: number;
+      message: string;
+    }>(`/admin/api/upstream-providers/${providerId}/batch-override`, payload);
+  }
+
   static async getProviderModel(
     providerId: number,
     modelId: string
@@ -498,8 +540,8 @@ export class AdminService {
       : null;
 
     const pricing = {
-      prompt: (data.input_cost as number) / 1000000,
-      completion: (data.output_cost as number) / 1000000,
+      prompt: data.input_cost as number,
+      completion: data.output_cost as number,
       request: (data.min_cost_per_request as number) || 0,
       image: 0,
       web_search: 0,
@@ -553,8 +595,8 @@ export class AdminService {
     const existingModel = await this.getModel(modelId, providerId);
 
     const pricing = {
-      prompt: (data.input_cost as number) / 1000000,
-      completion: (data.output_cost as number) / 1000000,
+      prompt: data.input_cost as number,
+      completion: data.output_cost as number,
       request: (data.min_cost_per_request as number) || 0,
       image: 0,
       web_search: 0,
@@ -926,6 +968,7 @@ export const TemporaryBalanceSchema = z.object({
   total_requests: z.number(),
   refund_address: z.string().nullable(),
   key_expiry_time: z.number().nullable(),
+  parent_key_hash: z.string().nullable().optional(),
 });
 
 export type TemporaryBalance = z.infer<typeof TemporaryBalanceSchema>;

ui/lib/api/services/auth.ts

@@ -3,28 +3,6 @@ import { apiClient } from '../client';
 import { ConfigurationService } from './configuration';
 import axios from 'axios';
 
-export const loginSchema = z.object({
-  username: z.string().optional(),
-  password: z.string().optional(),
-});
-
-export type LoginRequest = z.infer<typeof loginSchema>;
-
-export const loginResponseSchema = z.object({
-  id: z.string(),
-});
-
-export type LoginResponse = z.infer<typeof loginResponseSchema>;
-
-export async function login(data: LoginRequest): Promise<LoginResponse> {
-  try {
-    return await apiClient.post<LoginResponse>('/api/login', data);
-  } catch (error) {
-    console.error('Login error:', error);
-    throw error;
-  }
-}
-
 export const adminLoginSchema = z.object({
   password: z.string().min(1, 'Password is required'),
 });
@@ -91,40 +69,3 @@ export async function adminLogout(): Promise<void> {
     ConfigurationService.clearToken();
   }
 }
-
-export const registerSchema = z.object({
-  npub: z.string().min(10, { message: 'must have at least 10 character' }),
-  name: z.string().optional(),
-});
-
-export type RegisterRequest = z.infer<typeof registerSchema>;
-export type SchemaRegisterProps = z.infer<typeof registerSchema>;
-
-export const registerResponseSchema = z.object({
-  user_id: z.string(),
-  theme: z.string(),
-});
-
-export type RegisterResponse = z.infer<typeof registerResponseSchema>;
-
-export async function register(
-  data: RegisterRequest
-): Promise<RegisterResponse> {
-  try {
-    return await apiClient.post<RegisterResponse>('/api/register', data);
-  } catch (error) {
-    console.error('Registration error:', error);
-    throw error;
-  }
-}
-
-export const registerUser = register;
-
-export async function getUserSettings(): Promise<{ id: string }> {
-  try {
-    return await apiClient.get<{ id: string }>('/api/user/settings');
-  } catch (error) {
-    console.error('Error fetching user settings:', error);
-    throw error;
-  }
-}

ui/lib/api/services/wallet.ts

@@ -29,6 +29,28 @@ export type RedeemTokenResponse = z.infer<typeof RedeemTokenResponseSchema>;
 export type SendTokenRequest = z.infer<typeof SendTokenRequestSchema>;
 export type SendTokenResponse = z.infer<typeof SendTokenResponseSchema>;
 
+export interface BalanceDetail {
+  mint_url: string;
+  unit: string;
+  wallet_balance: number;
+  user_balance: number;
+  owner_balance: number;
+  error?: string;
+}
+
+export interface WithdrawResponse {
+  token: string;
+}
+
+export interface CreateChildKeyResponse {
+  api_keys: string[];
+  count: number;
+  cost_msats: number;
+  cost_sats: number;
+  parent_balance: number;
+  parent_balance_sats: number;
+}
+
 export class WalletService {
   static async redeemToken(token: string): Promise<RedeemTokenResponse> {
     try {
@@ -108,17 +130,84 @@ export class WalletService {
       throw error;
     }
   }
-}
 
-export interface BalanceDetail {
-  mint_url: string;
-  unit: string;
-  wallet_balance: number;
-  user_balance: number;
-  owner_balance: number;
-  error?: string;
-}
+  static async createChildKey(
+    baseUrl?: string,
+    apiKey?: string,
+    count: number = 1,
+    balanceLimit?: number,
+    balanceLimitReset?: string,
+    validityDate?: number
+  ): Promise<CreateChildKeyResponse> {
+    try {
+      if (baseUrl && apiKey) {
+        const response = await fetch(`${baseUrl}/v1/balance/child-key`, {
+          method: 'POST',
+          headers: {
+            'Content-Type': 'application/json',
+            Authorization: `Bearer ${apiKey}`,
+          },
+          body: JSON.stringify({
+            count,
+            balance_limit: balanceLimit,
+            balance_limit_reset: balanceLimitReset,
+            validity_date: validityDate,
+          }),
+        });
 
-export interface WithdrawResponse {
-  token: string;
+        if (!response.ok) {
+          const errorText = await response.text();
+          throw new Error(errorText || 'Failed to create child key');
+        }
+
+        return (await response.json()) as CreateChildKeyResponse;
+      }
+
+      return await apiClient.post<CreateChildKeyResponse>(
+        '/v1/balance/child-key',
+        {
+          count,
+          balance_limit: balanceLimit,
+          balance_limit_reset: balanceLimitReset,
+          validity_date: validityDate,
+        }
+      );
+    } catch (error) {
+      console.error('Error creating child key:', error);
+      throw error;
+    }
+  }
+
+  static async resetChildKeySpent(
+    baseUrl: string | undefined,
+    parentKey: string,
+    childKey: string
+  ): Promise<{ success: boolean; message: string }> {
+    try {
+      const url = baseUrl
+        ? `${baseUrl}/v1/balance/child-key/reset`
+        : '/v1/balance/child-key/reset';
+
+      const headers: Record<string, string> = {
+        'Content-Type': 'application/json',
+        Authorization: `Bearer ${parentKey}`,
+      };
+
+      const response = await fetch(url, {
+        method: 'POST',
+        headers,
+        body: JSON.stringify({ child_key: childKey }),
+      });
+
+      if (!response.ok) {
+        const errorText = await response.text();
+        throw new Error(errorText || 'Failed to reset child key');
+      }
+
+      return await response.json();
+    } catch (error) {
+      console.error('Error resetting child key:', error);
+      throw error;
+    }
+  }
 }