Merge branch 'dynamic-prices' into litellm

make sure to always emit sats cost

.env.example

@@ -14,6 +14,7 @@ UPSTREAM_API_KEY=your-upstream-api-key
 # HTTP_URL=https://api.mynode.com
 # ONION_URL=http://mynode.onion (auto fetched from compose)
 # RELAYS="wss://relay.damus.io,wss://relay.nostr.band,wss://eden.nostr.land,wss://relay.routstr.com"
+# ENABLE_ANALYTICS_SHARING=true
 # CASHU_MINTS="https://mint.minibits.cash/Bitcoin,https://mint.cubabitcoin.org,https://ecashmint.otrta.me"
 # RECEIVE_LN_ADDRESS=
 

.github/workflows/test.yml

@@ -67,7 +67,7 @@ jobs:
       - name: Setup Node.js
         uses: actions/setup-node@v4
         with:
-          node-version: "18"
+          node-version: "20"
           cache: "pnpm"
           cache-dependency-path: ui/pnpm-lock.yaml
 

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 +1,81 @@
 # 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
-
-- **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
-
-## Getting Started
-
-### Running the proxy using Docker
+### cURL
 
 ```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
+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"}]
+  }'
 ```
 
-### Development Requirements
+## Quick Start (Docker)
 
-- Python 3.11+
-- [uv](https://github.com/astral-sh/uv) package manager (used in development)
+If you are a node runner, start a Routstr Core instance using Docker Compose:
 
-### Installation
-
-```bash
-uv sync  # install dependencies
-```
-
-Create a `.env` file based on `.env.example` and fill in the required values:
+1. **Prepare your `.env`**:
+   ```bash
+   ADMIN_PASSWORD=mysecretpassword
+   NAME="My AI Node"
+   DESCRIPTION="Fast access to models"
+   NSEC=yournsec
+   RECEIVE_LN_ADDRESS=yourname@wallet.com
+   ```
+
+2. **Start the services**:
+   ```bash
+   docker compose up -d
+   ```
+
+3. **Configure**:
+   Open [http://localhost:8000/admin/](http://localhost:8000/admin/) to connect your AI providers and set pricing.
+
+For full instructions, see the **[Provider Quick Start Guide](https://docs.routstr.com/provider/quickstart/)**.
+
+## 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

@@ -360,6 +360,42 @@ POST /v1/wallet/create
 }
 ```
 
+### Get Key Information
+
+Get current balance, consumption data, and child keys for an API key.
+
+```http
+GET /v1/balance/info
+Authorization: Bearer sk-...
+```
+
+**Response:**
+
+```json
+{
+  "api_key": "sk-abc...",
+  "balance": 8500000,
+  "reserved": 0,
+  "is_child": false,
+  "parent_key": null,
+  "total_requests": 42,
+  "total_spent": 1500000,
+  "balance_limit": null,
+  "balance_limit_reset": null,
+  "validity_date": null,
+  "child_keys": [
+    {
+      "api_key": "sk-child1...",
+      "total_requests": 10,
+      "total_spent": 500000,
+      "balance_limit": 1000000,
+      "balance_limit_reset": "daily",
+      "validity_date": 1738000000
+    }
+  ]
+}
+```
+
 ### Check Balance
 
 Get current wallet balance.
@@ -563,4 +599,4 @@ 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 +1,150 @@
+# 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.
+
+---
+
+## Initial Setup (.env file)
+
+Before running your node, you should create a `.env` file in the project root. This file is used to bootstrap the initial configuration and store sensitive secrets.
+
+### Example .env
+
+```bash
+ADMIN_PASSWORD=your-secure-password
+
+# Node Identity
+NAME="My AI Node"
+DESCRIPTION="Fast access to models"
+
+# Lightning Payouts
+RECEIVE_LN_ADDRESS=yourname@wallet.com
+```
+
+### Setting the UI Password
+
+There are two ways to set or change your Admin Dashboard password:
+
+1.  **Via Environment Variable**: Set `ADMIN_PASSWORD` in your `.env` file before starting the container. This will be the password used for the first login.
+2.  **Via Dashboard**: Once logged in, go to **Settings** → **Security** to update your password. Dashboard settings override the `.env` file once saved.
+
+---
+
+## 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      |
+| **Share Analytics** | Publish aggregate usage stats to Nostr |
+
+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                 | —                                    |
+| `ENABLE_ANALYTICS_SHARING` | Enable usage analytics sharing to Nostr | `true`                         |
+| `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 +1,209 @@
+# 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 |
+| **Share Analytics** | Toggle publishing aggregate usage stats to Nostr |
+
+### 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 +1,190 @@
+# Deployment
+
+Production deployment guide for Routstr Provider nodes.
+
+## Docker Compose (Recommended)
+
+For production, use Docker Compose with persistent storage and optional Tor support.
+
+Use the included `compose.yml` for a flexible setup that handles both the UI and 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
+
+### Using Docker Compose
+The easiest way to build everything from source:
+
+```bash
+docker compose build
+```
+
+### Individual Components
+If you prefer building the node only (requires manual UI build first):
+
+```bash
+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 +1,124 @@
+# 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. Prepare Configuration
+
+Create a `.env` file in the root of the project to store your secrets:
+
+```bash
+# Initial Admin Password
+ADMIN_PASSWORD=mysecretpassword
+
+# Node Identity
+NAME="My AI Node"
+DESCRIPTION="Fast access to models"
+NSEC=yournsec
+
+# Lightning Payouts
+RECEIVE_LN_ADDRESS=yourname@wallet.com
+
+```
+
+## 2. Start the Node
+
+The recommended way to run Routstr is using Docker Compose, which handles the node, the UI, and optional services like Tor.
+
+```bash
+docker compose up -d
+```
+
+Verify it's running:
+
+```bash
+curl http://localhost:8000/v1/info
+```
+
+### Build from Source (Optional)
+
+If you've cloned the repository and want to build the images yourself:
+
+```bash
+docker compose build
+docker compose up -d
+```
+
+---
+
+## 3. Configure via Dashboard
+
+Open the **Admin Dashboard** at [http://localhost:8000/admin/](http://localhost:8000/admin/).
+
+!!! note "Login"
+Use the `ADMIN_PASSWORD` you defined in your `.env` file to log in. If you didn't set one, the dashboard will prompt you to set one on first visit.
+
+### 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

logs/.gitkeep

@@ -0,0 +1 @@
+

migrations/versions/02650cd6f028_add_routstr_fees_table.py

@@ -0,0 +1,32 @@
+"""add routstr_fees table
+
+Revision ID: 02650cd6f028
+Revises: c3d4e5f6a7b8
+Create Date: 2026-04-24 00:00:00.000000
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "02650cd6f028"
+down_revision = "c3d4e5f6a7b8"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "routstr_fees",
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("accumulated_msats", sa.Integer(), nullable=False, server_default="0"),
+        sa.Column("total_paid_msats", sa.Integer(), nullable=False, server_default="0"),
+        sa.Column("last_paid_at", sa.Integer(), nullable=True),
+        sa.PrimaryKeyConstraint("id"),
+    )
+    # Seed with a single row
+    op.execute("INSERT INTO routstr_fees (id, accumulated_msats, total_paid_msats) VALUES (1, 0, 0)")
+
+
+def downgrade() -> None:
+    op.drop_table("routstr_fees")

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/614c0a740e68_add_provider_settings_to_upstream_.py

@@ -0,0 +1,32 @@
+"""add provider_settings to upstream_providers
+
+Revision ID: 614c0a740e68
+Revises: 06f81c0fc88d
+Create Date: 2026-02-13 22:36:53.608737
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "614c0a740e68"
+down_revision = "06f81c0fc88d"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # Check if column exists before adding it
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    columns = [c["name"] for c in inspector.get_columns("upstream_providers")]
+
+    if "provider_settings" not in columns:
+        op.add_column(
+            "upstream_providers",
+            sa.Column("provider_settings", sa.Text(), nullable=True),
+        )
+
+
+def downgrade() -> None:
+    op.drop_column("upstream_providers", "provider_settings")

migrations/versions/6d2fa295fa43_add_provider_fee_schedules.py

@@ -0,0 +1,57 @@
+"""add provider_fee_schedules and provider_fee_default to upstream_providers
+
+Revision ID: 6d2fa295fa43
+Revises: cli_tokens_001
+Create Date: 2026-04-28 00:00:00.000000
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "6d2fa295fa43"
+down_revision = "cli_tokens_001"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    columns = {c["name"] for c in inspector.get_columns("upstream_providers")}
+
+    if "provider_fee_default" not in columns:
+        op.add_column(
+            "upstream_providers",
+            sa.Column(
+                "provider_fee_default",
+                sa.Float(),
+                nullable=False,
+                server_default="1.01",
+            ),
+        )
+        # Preserve any custom per-provider fees by copying from provider_fee.
+        op.execute(
+            "UPDATE upstream_providers "
+            "SET provider_fee_default = provider_fee "
+            "WHERE provider_fee IS NOT NULL"
+        )
+
+    if "provider_fee_schedules" not in columns:
+        op.add_column(
+            "upstream_providers",
+            sa.Column("provider_fee_schedules", sa.Text(), nullable=True),
+        )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    columns = {c["name"] for c in inspector.get_columns("upstream_providers")}
+
+    if "provider_fee_schedules" in columns:
+        op.drop_column("upstream_providers", "provider_fee_schedules")
+    if "provider_fee_default" in columns:
+        op.drop_column("upstream_providers", "provider_fee_default")

migrations/versions/a776ca70e5fe_add_cashu_refunds_table.py

@@ -0,0 +1,42 @@
+"""add cashu_transactions table
+
+Revision ID: a776ca70e5fe
+Revises: 614c0a740e68
+Create Date: 2026-03-11 22:00:01.554762
+"""
+
+import sqlalchemy as sa
+import sqlmodel
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "a776ca70e5fe"
+down_revision = "614c0a740e68"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "cashu_transactions",
+        sa.Column("id", sqlmodel.sql.sqltypes.AutoString(), nullable=False),
+        sa.Column("token", sqlmodel.sql.sqltypes.AutoString(), nullable=False),
+        sa.Column("amount", sa.Integer(), nullable=False),
+        sa.Column("unit", sqlmodel.sql.sqltypes.AutoString(), nullable=False),
+        sa.Column("mint_url", sqlmodel.sql.sqltypes.AutoString(), nullable=True),
+        sa.Column(
+            "type",
+            sqlmodel.sql.sqltypes.AutoString(),
+            nullable=False,
+            server_default="out",
+        ),
+        sa.Column("request_id", sqlmodel.sql.sqltypes.AutoString(), nullable=True),
+        sa.Column("created_at", sa.Integer(), nullable=False),
+        sa.Column("collected", sa.Boolean(), nullable=False),
+        sa.Column("swept", sa.Boolean(), nullable=False),
+        sa.PrimaryKeyConstraint("id"),
+    )
+
+
+def downgrade() -> None:
+    op.drop_table("cashu_transactions")

migrations/versions/b1c2d3e4f5a6_add_upstream_model_id_to_models.py

@@ -0,0 +1,33 @@
+"""add forwarded_model_id to models
+
+Revision ID: b1c2d3e4f5a6
+Revises: a776ca70e5fe
+Create Date: 2026-04-05 00:00:00.000000
+"""
+
+import sqlalchemy as sa
+import sqlmodel
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "b1c2d3e4f5a6"
+down_revision = "a776ca70e5fe"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "models",
+        sa.Column(
+            "forwarded_model_id",
+            sqlmodel.sql.sqltypes.AutoString(),
+            nullable=True,
+        ),
+    )
+    # Backfill: set forwarded_model_id = id for all existing rows
+    op.execute("UPDATE models SET forwarded_model_id = id WHERE forwarded_model_id IS NULL")
+
+
+def downgrade() -> None:
+    op.drop_column("models", "forwarded_model_id")

migrations/versions/c3d4e5f6a7b8_add_source_to_cashu_transactions.py

@@ -0,0 +1,36 @@
+"""add source to cashu_transactions
+
+Revision ID: c3d4e5f6a7b8
+Revises: b1c2d3e4f5a6
+Create Date: 2026-04-10 00:00:00.000000
+"""
+
+import sqlalchemy as sa
+import sqlmodel
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "c3d4e5f6a7b8"
+down_revision = "b1c2d3e4f5a6"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    columns = [col["name"] for col in inspector.get_columns("cashu_transactions")]
+    if "source" not in columns:
+        op.add_column(
+            "cashu_transactions",
+            sa.Column(
+                "source",
+                sqlmodel.sql.sqltypes.AutoString(),
+                nullable=False,
+                server_default="x-cashu",
+            ),
+        )
+
+
+def downgrade() -> None:
+    op.drop_column("cashu_transactions", "source")

migrations/versions/cli_tokens_table.py

@@ -0,0 +1,34 @@
+"""add cli_tokens table
+
+Revision ID: cli_tokens_001
+Revises: e8f9a0b1c2d3
+Create Date: 2026-04-25 00:00:00.000000
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "cli_tokens_001"
+down_revision = "e8f9a0b1c2d3"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "cli_tokens",
+        sa.Column("id", sa.String(), primary_key=True, nullable=False),
+        sa.Column("token", sa.String(), nullable=False, unique=True),
+        sa.Column("name", sa.String(), nullable=False),
+        sa.Column("created_at", sa.Integer(), nullable=False),
+        sa.Column("last_used_at", sa.Integer(), nullable=True),
+        sa.Column("expires_at", sa.Integer(), nullable=True),
+    )
+    op.create_index("ix_cli_tokens_token", "cli_tokens", ["token"], unique=True)
+
+
+def downgrade() -> None:
+    op.drop_index("ix_cli_tokens_token", table_name="cli_tokens")
+    op.drop_table("cli_tokens")

migrations/versions/d4e5f6a7b8c9_add_api_key_to_cashu_transactions.py

@@ -0,0 +1,46 @@
+"""add api key link to cashu_transactions
+
+Revision ID: d4e5f6a7b8c9
+Revises: c3d4e5f6a7b8
+Create Date: 2026-04-20 00:00:00.000000
+"""
+
+import sqlalchemy as sa
+import sqlmodel
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "d4e5f6a7b8c9"
+down_revision = "c3d4e5f6a7b8"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    columns = [col["name"] for col in inspector.get_columns("cashu_transactions")]
+    indexes = {index["name"] for index in inspector.get_indexes("cashu_transactions")}
+
+    if "api_key_hashed_key" not in columns:
+        op.add_column(
+            "cashu_transactions",
+            sa.Column(
+                "api_key_hashed_key",
+                sqlmodel.sql.sqltypes.AutoString(),
+                nullable=True,
+            ),
+        )
+
+    if "ix_cashu_transactions_api_key_hashed_key" not in indexes:
+        op.create_index(
+            "ix_cashu_transactions_api_key_hashed_key",
+            "cashu_transactions",
+            ["api_key_hashed_key"],
+            unique=False,
+        )
+
+
+def downgrade() -> None:
+    op.drop_index("ix_cashu_transactions_api_key_hashed_key", table_name="cashu_transactions")
+    op.drop_column("cashu_transactions", "api_key_hashed_key")

migrations/versions/e8f9a0b1c2d3_merge_heads.py

@@ -0,0 +1,20 @@
+"""merge heads: routstr_fees + api_key_to_cashu_transactions
+
+Revision ID: e8f9a0b1c2d3
+Revises: 02650cd6f028, d4e5f6a7b8c9
+Create Date: 2026-04-24 00:00:00.000000
+"""
+
+# revision identifiers, used by Alembic.
+revision = "e8f9a0b1c2d3"
+down_revision = ("02650cd6f028", "d4e5f6a7b8c9")
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    pass
+
+
+def downgrade() -> None:
+    pass

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 = "0.3.0"
+version = "0.4.3"
 description = "Payment proxy for your LLM endpoint using cashu and nostr."
 readme = "README.md"
 requires-python = ">=3.11"
@@ -13,7 +13,7 @@ dependencies = [
     "greenlet>=3.2.1",
     "alembic>=1.13",
     "python-json-logger>=2.0.0",
-    "cashu",
+    "cashu>=0.20",
     "secp256k1",
     "marshmallow>=3.13,<4.0",
     "websockets>=12.0",
@@ -21,6 +21,7 @@ dependencies = [
     "mdurl==0.1.2",
     "pillow>=10",
     "openai>=1.98.0",
+    "litellm>=1.55.0",
 ]
 
 [dependency-groups]

routstr/algorithm.py

@@ -118,12 +118,32 @@ def create_model_mappings(
 
     candidates: dict[str, list[tuple["Model", "BaseUpstreamProvider"]]] = {}
     unique_models: dict[str, "Model"] = {}
+    seen_model_provider: set[tuple[str, str]] = set()
+
+    providers_by_db_id: dict[int, "BaseUpstreamProvider"] = {}
+    for upstream in upstreams:
+        db_id = getattr(upstream, "db_id", None)
+        if isinstance(db_id, int):
+            providers_by_db_id[db_id] = upstream
+
+    # Group upstreams by URL and keep only the one with the lowest fee for each URL
+    upstreams_by_url: dict[str, list["BaseUpstreamProvider"]] = {}
+    for upstream in upstreams:
+        url = getattr(upstream, "base_url", "")
+        if url not in upstreams_by_url:
+            upstreams_by_url[url] = []
+        upstreams_by_url[url].append(upstream)
+
+    filtered_upstreams: list["BaseUpstreamProvider"] = []
+    for providers in upstreams_by_url.values():
+        best_provider = min(providers, key=lambda p: p.provider_fee)
+        filtered_upstreams.append(best_provider)
 
     # Separate OpenRouter from other providers
     openrouter: "BaseUpstreamProvider" | None = None
     other_upstreams: list["BaseUpstreamProvider"] = []
 
-    for upstream in upstreams:
+    for upstream in filtered_upstreams:
         base_url = getattr(upstream, "base_url", "")
         if base_url == "https://openrouter.ai/api/v1":
             openrouter = upstream
@@ -134,6 +154,16 @@ 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 get_provider_identity(upstream: "BaseUpstreamProvider") -> str:
+        """Get a stable provider identity used for deduplication."""
+        db_id = getattr(upstream, "db_id", None)
+        if isinstance(db_id, int):
+            return f"db:{db_id}"
+
+        provider_type = str(getattr(upstream, "provider_type", "") or "").lower()
+        base_url = str(getattr(upstream, "base_url", "") or "").lower()
+        return f"{provider_type}|{base_url}"
+
     def _add_candidate(
         alias: str, model: "Model", provider: "BaseUpstreamProvider"
     ) -> None:
@@ -148,6 +178,7 @@ def create_model_mappings(
     ) -> None:
         """Process all models from a given provider."""
         upstream_prefix = getattr(upstream, "upstream_name", None)
+        provider_key = get_provider_identity(upstream)
 
         for model in upstream.get_cached_models():
             if not model.enabled or model.id in disabled_model_ids:
@@ -186,9 +217,14 @@ def create_model_mappings(
                 if prefixed_id not in aliases:
                     aliases.append(prefixed_id)
 
+            # Register forwarded_model_id as a routable alias
+            if model_to_use.forwarded_model_id and model_to_use.forwarded_model_id not in aliases:
+                aliases.append(model_to_use.forwarded_model_id)
+
             # Try to set each alias
             for alias in aliases:
                 _add_candidate(alias, model_to_use, upstream)
+            seen_model_provider.add((model_to_use.id.lower(), provider_key))
 
     # Process non-OpenRouter providers first
     for upstream in other_upstreams:
@@ -198,6 +234,89 @@ def create_model_mappings(
     if openrouter:
         process_provider_models(openrouter, is_openrouter=True)
 
+    # Include enabled DB overrides even when provider discovery misses models.
+    # This is important for deployment-based providers like Azure.
+    for model_id, override_data in overrides_by_id.items():
+        if model_id in disabled_model_ids:
+            continue
+        override_row, provider_fee = override_data
+        upstream_provider_id = getattr(override_row, "upstream_provider_id", None)
+        if not isinstance(upstream_provider_id, int):
+            continue
+
+        upstream_for_override = providers_by_db_id.get(upstream_provider_id)
+        if upstream_for_override is None:
+            continue
+
+        provider_key = get_provider_identity(upstream_for_override)
+        dedupe_key = (model_id.lower(), provider_key)
+        if dedupe_key in seen_model_provider:
+            continue
+
+        try:
+            model_to_use = _row_to_model(
+                override_row, apply_provider_fee=True, provider_fee=provider_fee
+            )
+        except Exception as exc:
+            logger.warning(
+                "Skipping invalid model override while building model mappings",
+                extra={
+                    "model_id": model_id,
+                    "upstream_provider_id": upstream_provider_id,
+                    "error": str(exc),
+                    "error_type": type(exc).__name__,
+                },
+            )
+            continue
+        if not model_to_use.enabled:
+            continue
+
+        base_id = get_base_model_id(model_to_use.id)
+        is_openrouter = (
+            getattr(upstream_for_override, "base_url", "")
+            == "https://openrouter.ai/api/v1"
+        )
+        if not is_openrouter or base_id not in unique_models:
+            unique_model = model_to_use.copy(
+                update={
+                    "id": base_id,
+                    "upstream_provider_id": upstream_for_override.provider_type,
+                }
+            )
+            unique_models[base_id] = unique_model
+
+        try:
+            aliases = resolve_model_alias(
+                model_to_use.id,
+                model_to_use.canonical_slug,
+                alias_ids=model_to_use.alias_ids,
+            )
+        except Exception as exc:
+            logger.warning(
+                "Skipping model aliases for invalid override model",
+                extra={
+                    "model_id": model_id,
+                    "upstream_provider_id": upstream_provider_id,
+                    "error": str(exc),
+                    "error_type": type(exc).__name__,
+                },
+            )
+            continue
+
+        upstream_prefix = getattr(upstream_for_override, "upstream_name", None)
+        if upstream_prefix and "/" not in model_to_use.id:
+            prefixed_id = f"{upstream_prefix}/{model_to_use.id}"
+            if prefixed_id not in aliases:
+                aliases.append(prefixed_id)
+
+        # Register forwarded_model_id as a routable alias
+        if model_to_use.forwarded_model_id and model_to_use.forwarded_model_id not in aliases:
+            aliases.append(model_to_use.forwarded_model_id)
+
+        for alias in aliases:
+            _add_candidate(alias, model_to_use, upstream_for_override)
+        seen_model_provider.add(dedupe_key)
+
     # Sort candidates and build final maps
     model_instances: dict[str, "Model"] = {}
     provider_map: dict[str, list["BaseUpstreamProvider"]] = {}

routstr/auth.py

@@ -1,13 +1,18 @@
+import asyncio
 import hashlib
 import math
+import random
+import time
+from datetime import datetime
 from typing import Optional
 
 from fastapi import HTTPException
+from sqlalchemy import case
 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
+from .core.db import ApiKey, AsyncSession, accumulate_routstr_fee
 from .core.settings import settings
 from .payment.cost_calculation import (
     CostData,
@@ -18,22 +23,73 @@ from .payment.cost_calculation import (
 from .wallet import credit_balance, deserialize_token_from_string
 
 logger = get_logger(__name__)
+payments_logger = get_logger("routstr.payments")
+
+# Routstr platform fee constants
+ROUTSTR_FEE_PERCENT: float = 2.1
+ROUTSTR_LN_ADDRESS: str = "npub130mznv74rxs032peqym6g3wqavh472623mt3z5w73xq9r6qqdufs7ql29s@npub.cash"
+ROUTSTR_FEE_PAYOUT_INTERVAL_SECONDS: int = 900
+ROUTSTR_FEE_DEFAULT_PAYOUT: int = 200
 
 # TODO: implement prepaid api key (not like it was before)
 # PREPAID_API_KEY = os.environ.get("PREPAID_API_KEY", None)
 # 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 +99,7 @@ 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 +154,50 @@ 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 +253,19 @@ 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(
@@ -242,6 +356,8 @@ async def validate_bearer_key(
             )
 
             return new_key
+        except HTTPException:
+            raise
         except Exception as e:
             logger.error(
                 "Cashu token redemption failed",
@@ -264,12 +380,12 @@ async def validate_bearer_key(
                 },
             )
 
+    key_preview = bearer_key[:10] + "..." if len(bearer_key) > 10 else bearer_key
     logger.error(
-        "Invalid API key format",
+        f"Invalid API key format: preview={key_preview!r} length={len(bearer_key)} "
+        f"(expected 'sk-...' or 'cashu...' token)",
         extra={
-            "key_preview": bearer_key[:10] + "..."
-            if len(bearer_key) > 10
-            else bearer_key,
+            "key_preview": key_preview,
             "key_length": len(bearer_key),
         },
     )
@@ -278,7 +394,7 @@ async def validate_bearer_key(
         status_code=401,
         detail={
             "error": {
-                "message": "Invalid API key",
+                "message": "Invalid API key format. Expected an 'sk-...' API key or a 'cashu...' token.",
                 "type": "invalid_request_error",
                 "code": "invalid_api_key",
             }
@@ -311,6 +427,8 @@ 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)
 
@@ -349,6 +467,57 @@ async def pay_for_request(
             },
         )
 
+    # 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={
@@ -371,12 +540,15 @@ async def pay_for_request(
     )
     result = await session.exec(stmt)  # type: ignore[call-overload]
 
-    # Also increment total_requests on the child key if it's different
+    # 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)
+            .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]
 
@@ -420,18 +592,33 @@ async def pay_for_request(
             "total_requests": billing_key.total_requests,
         },
     )
+    payments_logger.info(
+        "RESERVE",
+        extra={
+            "event": "reserve",
+            "key_hash": key.hashed_key[:8] + "...",
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
+            "cost_reserved": cost_per_request,
+            "balance": billing_key.balance,
+            "reserved_balance": billing_key.reserved_balance,
+            "total_spent": billing_key.total_spent,
+        },
+    )
 
     return cost_per_request
 
 
 async def revert_pay_for_request(
     key: ApiKey, session: AsyncSession, cost_per_request: int
-) -> None:
+) -> bool:
+    """Revert a previously reserved payment. Returns True if revert succeeded,
+    False if the reservation was already released (prevents negative reserved_balance)."""
     billing_key = await get_billing_key(key, session)
 
     stmt = (
         update(ApiKey)
         .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
+        .where(col(ApiKey.reserved_balance) >= cost_per_request)
         .values(
             reserved_balance=col(ApiKey.reserved_balance) - cost_per_request,
             total_requests=col(ApiKey.total_requests) - 1,
@@ -440,19 +627,23 @@ async def revert_pay_for_request(
 
     result = await session.exec(stmt)  # type: ignore[call-overload]
 
-    # Also decrement total_requests on the child key if it's different
+    # 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)
+            .where(col(ApiKey.reserved_balance) >= cost_per_request)
+            .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",
+        logger.warning(
+            "Revert skipped - reservation already released (no-op to prevent negative reserved_balance)",
             extra={
                 "key_hash": key.hashed_key[:8] + "...",
                 "billing_key_hash": billing_key.hashed_key[:8] + "...",
@@ -460,19 +651,22 @@ async def revert_pay_for_request(
                 "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. {billing_key.balance} available.",
-                    "type": "payment_error",
-                    "code": "payment_error",
-                }
-            },
-        )
+        return False
     await session.refresh(billing_key)
     if billing_key.hashed_key != key.hashed_key:
         await session.refresh(key)
+    payments_logger.info(
+        "REVERT",
+        extra={
+            "event": "revert",
+            "key_hash": key.hashed_key[:8] + "...",
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
+            "cost_reverted": cost_per_request,
+            "balance": billing_key.balance,
+            "reserved_balance": billing_key.reserved_balance,
+        },
+    )
+    return True
 
 
 async def adjust_payment_for_tokens(
@@ -504,20 +698,45 @@ async def adjust_payment_for_tokens(
             release_stmt = (
                 update(ApiKey)
                 .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
+                .where(col(ApiKey.reserved_balance) >= deducted_max_cost)
                 .values(
                     reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost
                 )
             )
-            await session.exec(release_stmt)  # type: ignore[call-overload]
+            result = 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)
+                    .where(col(ApiKey.reserved_balance) >= deducted_max_cost)
+                    .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,
-                },
-            )
+            if result.rowcount == 0:  # type: ignore[union-attr]
+                logger.warning(
+                    "Release reservation skipped - already released (no-op to prevent negative reserved_balance)",
+                    extra={
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                        "deducted_max_cost": deducted_max_cost,
+                    },
+                )
+            else:
+                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",
@@ -528,6 +747,17 @@ async def adjust_payment_for_tokens(
                 },
             )
 
+    async def _accumulate_fee(total_cost_msats: int) -> None:
+        if total_cost_msats > 0 and ROUTSTR_FEE_PERCENT > 0:
+            fee_msats = math.ceil(total_cost_msats * ROUTSTR_FEE_PERCENT / 100)
+            try:
+                await accumulate_routstr_fee(session, fee_msats)
+            except Exception as e:
+                logger.warning(
+                    "Failed to accumulate Routstr fee",
+                    extra={"error": str(e), "fee_msats": fee_msats},
+                )
+
     match await calculate_cost(response_data, deducted_max_cost, session):
         case MaxCostData() as cost:
             logger.debug(
@@ -540,23 +770,52 @@ async def adjust_payment_for_tokens(
                 },
             )
             # Finalize by releasing reservation and charging max cost
+            if billing_key.reserved_balance < deducted_max_cost:
+                logger.error(
+                    "reserved_balance below deducted_max_cost before MaxCost finalization — clamping to 0",
+                    extra={
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                        "reserved_balance": billing_key.reserved_balance,
+                        "deducted_max_cost": deducted_max_cost,
+                        "total_cost_msats": cost.total_msats,
+                        "balance": billing_key.balance,
+                        "total_spent": billing_key.total_spent,
+                        "model": model,
+                    },
+                )
+
+            safe_reserved = case(
+                (col(ApiKey.reserved_balance) >= deducted_max_cost,
+                 col(ApiKey.reserved_balance) - deducted_max_cost),
+                else_=0,
+            )
+
             finalize_stmt = (
                 update(ApiKey)
                 .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                 .values(
-                    reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost,
+                    reserved_balance=safe_reserved,
                     balance=col(ApiKey.balance) - cost.total_msats,
                     total_spent=col(ApiKey.total_spent) + cost.total_msats,
                 )
             )
             result = await session.exec(finalize_stmt)  # type: ignore[call-overload]
 
-            # Also update total_spent on the child key if it's different
+            # Also update total_spent and reserved_balance on the child key if it's different
             if billing_key.hashed_key != key.hashed_key:
+                child_safe_reserved = case(
+                    (col(ApiKey.reserved_balance) >= deducted_max_cost,
+                     col(ApiKey.reserved_balance) - deducted_max_cost),
+                    else_=0,
+                )
                 child_stmt = (
                     update(ApiKey)
                     .where(col(ApiKey.hashed_key) == key.hashed_key)
-                    .values(total_spent=col(ApiKey.total_spent) + cost.total_msats)
+                    .values(
+                        total_spent=col(ApiKey.total_spent) + cost.total_msats,
+                        reserved_balance=child_safe_reserved,
+                    )
                 )
                 await session.exec(child_stmt)  # type: ignore[call-overload]
 
@@ -584,10 +843,30 @@ async def adjust_payment_for_tokens(
                         "key_hash": key.hashed_key[:8] + "...",
                         "billing_key_hash": billing_key.hashed_key[:8] + "...",
                         "charged_amount": cost.total_msats,
+                        "input_tokens": cost.input_tokens,
+                        "output_tokens": cost.output_tokens,
                         "new_balance": billing_key.balance,
                         "model": model,
                     },
                 )
+                await _accumulate_fee(cost.total_msats)
+                payments_logger.info(
+                    "FINALIZE",
+                    extra={
+                        "event": "finalize",
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                        "model": model,
+                        "cost_reserved": deducted_max_cost,
+                        "cost_charged": cost.total_msats,
+                        "input_tokens": cost.input_tokens,
+                        "output_tokens": cost.output_tokens,
+                        "balance": billing_key.balance,
+                        "reserved_balance": billing_key.reserved_balance,
+                        "total_spent": billing_key.total_spent,
+                        "finalize_type": "max_cost",
+                    },
+                )
             return cost.dict()
 
         case CostData() as cost:
@@ -607,6 +886,8 @@ async def adjust_payment_for_tokens(
                     "cost_difference": cost_difference,
                     "input_msats": cost.input_msats,
                     "output_msats": cost.output_msats,
+                    "input_tokens": cost.input_tokens,
+                    "output_tokens": cost.output_tokens,
                 },
             )
 
@@ -619,24 +900,52 @@ async def adjust_payment_for_tokens(
                         "model": model,
                     },
                 )
+                if billing_key.reserved_balance < deducted_max_cost:
+                    logger.error(
+                        "reserved_balance below deducted_max_cost on exact-cost finalization — clamping to 0",
+                        extra={
+                            "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                            "reserved_balance": billing_key.reserved_balance,
+                            "deducted_max_cost": deducted_max_cost,
+                            "total_cost_msats": total_cost_msats,
+                            "balance": billing_key.balance,
+                            "total_spent": billing_key.total_spent,
+                            "model": model,
+                        },
+                    )
+
+                exact_safe_reserved = case(
+                    (col(ApiKey.reserved_balance) >= deducted_max_cost,
+                     col(ApiKey.reserved_balance) - deducted_max_cost),
+                    else_=0,
+                )
+
                 finalize_stmt = (
                     update(ApiKey)
                     .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                     .values(
-                        reserved_balance=col(ApiKey.reserved_balance)
-                        - deducted_max_cost,
+                        reserved_balance=exact_safe_reserved,
                         balance=col(ApiKey.balance) - total_cost_msats,
                         total_spent=col(ApiKey.total_spent) + total_cost_msats,
                     )
                 )
                 await session.exec(finalize_stmt)  # type: ignore[call-overload]
 
-                # Also update total_spent on the child key if it's different
+                # Also update total_spent and reserved_balance on the child key if it's different
                 if billing_key.hashed_key != key.hashed_key:
+                    child_exact_safe_reserved = case(
+                        (col(ApiKey.reserved_balance) >= deducted_max_cost,
+                         col(ApiKey.reserved_balance) - deducted_max_cost),
+                        else_=0,
+                    )
                     child_stmt = (
                         update(ApiKey)
                         .where(col(ApiKey.hashed_key) == key.hashed_key)
-                        .values(total_spent=col(ApiKey.total_spent) + total_cost_msats)
+                        .values(
+                            total_spent=col(ApiKey.total_spent) + total_cost_msats,
+                            reserved_balance=child_exact_safe_reserved,
+                        )
                     )
                     await session.exec(child_stmt)  # type: ignore[call-overload]
 
@@ -644,52 +953,67 @@ async def adjust_payment_for_tokens(
                 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???
-            if cost_difference > 0:
-                # Need to charge more than reserved, finalize by releasing reservation and charging total
-                logger.info(
-                    "Additional charge required for token usage",
+                await _accumulate_fee(total_cost_msats)
+                payments_logger.info(
+                    "FINALIZE",
                     extra={
+                        "event": "finalize",
                         "key_hash": key.hashed_key[:8] + "...",
                         "billing_key_hash": billing_key.hashed_key[:8] + "...",
-                        "additional_charge": cost_difference,
-                        "current_balance": billing_key.balance,
-                        "sufficient_balance": billing_key.balance >= cost_difference,
                         "model": model,
+                        "cost_reserved": deducted_max_cost,
+                        "cost_charged": total_cost_msats,
+                        "input_tokens": cost.input_tokens,
+                        "output_tokens": cost.output_tokens,
+                        "balance": billing_key.balance,
+                        "reserved_balance": billing_key.reserved_balance,
+                        "total_spent": billing_key.total_spent,
+                        "finalize_type": "exact",
                     },
                 )
+                return cost.dict()
+
+            # actual cost exceeded discounted reservation (due to tolerance_percentage)
+            if cost_difference > 0:
+                # Always release the reservation and charge min(actual_cost, balance).
+                # Using a CASE expression makes this a single atomic UPDATE — no
+                # multi-level fallback needed and balance can never go negative.
+                chargeable = case(
+                    (col(ApiKey.balance) >= total_cost_msats, total_cost_msats),
+                    else_=col(ApiKey.balance),
+                )
 
                 finalize_stmt = (
                     update(ApiKey)
                     .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
+                    .where(col(ApiKey.reserved_balance) >= deducted_max_cost)
                     .values(
-                        reserved_balance=col(ApiKey.reserved_balance)
-                        - deducted_max_cost,
-                        balance=col(ApiKey.balance) - total_cost_msats,
-                        total_spent=col(ApiKey.total_spent) + total_cost_msats,
+                        reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost,
+                        balance=col(ApiKey.balance) - chargeable,
+                        total_spent=col(ApiKey.total_spent) + chargeable,
                     )
                 )
                 result = await session.exec(finalize_stmt)  # type: ignore[call-overload]
 
-                # Also update total_spent 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)
+                        .where(col(ApiKey.reserved_balance) >= deducted_max_cost)
+                        .values(
+                            reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost,
+                            total_spent=col(ApiKey.total_spent) + min(billing_key.balance, total_cost_msats),
+                        )
                     )
                     await session.exec(child_stmt)  # type: ignore[call-overload]
 
                 await session.commit()
 
                 if result.rowcount:
-                    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
                     logger.info(
                         "Finalized payment with additional charge",
                         extra={
@@ -700,9 +1024,29 @@ async def adjust_payment_for_tokens(
                             "model": model,
                         },
                     )
+                    await _accumulate_fee(total_cost_msats)
+                    payments_logger.info(
+                        "FINALIZE",
+                        extra={
+                            "event": "finalize",
+                            "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                            "model": model,
+                            "cost_reserved": deducted_max_cost,
+                            "cost_charged": total_cost_msats,
+                            "input_tokens": cost.input_tokens,
+                            "output_tokens": cost.output_tokens,
+                            "balance": billing_key.balance,
+                            "reserved_balance": billing_key.reserved_balance,
+                            "total_spent": billing_key.total_spent,
+                            "finalize_type": "overrun",
+                        },
+                    )
                 else:
+                    # Guard fired: reservation was already released by a concurrent
+                    # finalization for this key. Nothing left to do.
                     logger.warning(
-                        "Failed to finalize additional charge - releasing reservation",
+                        "Finalization skipped - reservation already released",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
                             "billing_key_hash": billing_key.hashed_key[:8] + "...",
@@ -710,7 +1054,6 @@ async def adjust_payment_for_tokens(
                             "model": model,
                         },
                     )
-                    await release_reservation_only()
             else:
                 # Refund some of the base cost
                 refund = abs(cost_difference)
@@ -725,24 +1068,53 @@ async def adjust_payment_for_tokens(
                     },
                 )
 
+                if billing_key.reserved_balance < deducted_max_cost:
+                    logger.error(
+                        "reserved_balance below deducted_max_cost on refund finalization — clamping to 0",
+                        extra={
+                            "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                            "reserved_balance": billing_key.reserved_balance,
+                            "deducted_max_cost": deducted_max_cost,
+                            "total_cost_msats": total_cost_msats,
+                            "refund_amount": refund,
+                            "balance": billing_key.balance,
+                            "total_spent": billing_key.total_spent,
+                            "model": model,
+                        },
+                    )
+
+                refund_safe_reserved = case(
+                    (col(ApiKey.reserved_balance) >= deducted_max_cost,
+                     col(ApiKey.reserved_balance) - deducted_max_cost),
+                    else_=0,
+                )
+
                 refund_stmt = (
                     update(ApiKey)
                     .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                     .values(
-                        reserved_balance=col(ApiKey.reserved_balance)
-                        - deducted_max_cost,
+                        reserved_balance=refund_safe_reserved,
                         balance=col(ApiKey.balance) - total_cost_msats,
                         total_spent=col(ApiKey.total_spent) + total_cost_msats,
                     )
                 )
                 result = await session.exec(refund_stmt)  # type: ignore[call-overload]
 
-                # Also update total_spent on the child key if it's different
+                # Also update total_spent and reserved_balance on the child key if it's different
                 if billing_key.hashed_key != key.hashed_key:
+                    child_refund_safe_reserved = case(
+                        (col(ApiKey.reserved_balance) >= deducted_max_cost,
+                         col(ApiKey.reserved_balance) - deducted_max_cost),
+                        else_=0,
+                    )
                     child_stmt = (
                         update(ApiKey)
                         .where(col(ApiKey.hashed_key) == key.hashed_key)
-                        .values(total_spent=col(ApiKey.total_spent) + total_cost_msats)
+                        .values(
+                            total_spent=col(ApiKey.total_spent) + total_cost_msats,
+                            reserved_balance=child_refund_safe_reserved,
+                        )
                     )
                     await session.exec(child_stmt)  # type: ignore[call-overload]
 
@@ -778,6 +1150,25 @@ async def adjust_payment_for_tokens(
                             "model": model,
                         },
                     )
+                    await _accumulate_fee(total_cost_msats)
+                    payments_logger.info(
+                        "FINALIZE",
+                        extra={
+                            "event": "finalize",
+                            "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                            "model": model,
+                            "cost_reserved": deducted_max_cost,
+                            "cost_charged": total_cost_msats,
+                            "refunded": refund,
+                            "input_tokens": cost.input_tokens,
+                            "output_tokens": cost.output_tokens,
+                            "balance": billing_key.balance,
+                            "reserved_balance": billing_key.reserved_balance,
+                            "total_spent": billing_key.total_spent,
+                            "finalize_type": "refund",
+                        },
+                    )
 
             return cost.dict()
 
@@ -803,15 +1194,67 @@ async def adjust_payment_for_tokens(
                     }
                 },
             )
-    # 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,
-    }
+    # All calculate_cost variants are handled above.
+    raise AssertionError("Unreachable: unhandled calculate_cost result")
+
+
+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

@@ -1,13 +1,22 @@
 import asyncio
 import hashlib
+import time
 from time import monotonic
 from typing import Annotated, NoReturn
 
 from fastapi import APIRouter, Depends, Header, HTTPException
+from fastapi.responses import JSONResponse
 from pydantic import BaseModel
+from sqlmodel import col, select, update
 
-from .auth import validate_bearer_key
-from .core.db import ApiKey, AsyncSession, get_session
+from .auth import get_billing_key, validate_bearer_key
+from .core.db import (
+    ApiKey,
+    AsyncSession,
+    CashuTransaction,
+    get_session,
+    store_cashu_transaction,
+)
 from .core.logging import get_logger
 from .core.settings import settings
 from .lightning import lightning_router
@@ -33,10 +42,8 @@ async def get_key_from_header(
 
 
 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 {
+    info = {
         "api_key": "sk-" + key.hashed_key,
         "balance": billing_key.balance,
         "reserved": billing_key.reserved_balance,
@@ -44,8 +51,31 @@ async def get_balance_info(key: ApiKey, session: AsyncSession) -> dict:
         "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,
     }
 
+    if not key.parent_key_hash:
+        # Fetch child keys if this is a parent key
+        statement = select(ApiKey).where(ApiKey.parent_key_hash == key.hashed_key)
+        results = await session.exec(statement)
+        child_keys = results.all()
+        if child_keys:
+            info["child_keys"] = [
+                {
+                    "api_key": "sk-" + ck.hashed_key,
+                    "total_requests": ck.total_requests,
+                    "total_spent": ck.total_spent,
+                    "balance_limit": ck.balance_limit,
+                    "balance_limit_reset": ck.balance_limit_reset,
+                    "validity_date": ck.validity_date,
+                }
+                for ck in child_keys
+            ]
+
+    return info
+
 
 # TODO: remove this endpoint when frontend is updated
 @router.get("/", include_in_schema=False)
@@ -70,9 +100,24 @@ async def account_info(
 
 @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,
@@ -98,8 +143,6 @@ 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:
@@ -118,9 +161,23 @@ async def topup_wallet_endpoint(
             raise HTTPException(status_code=400, detail="Token already spent")
         elif "invalid" in error_msg.lower() or "decode" in error_msg.lower():
             raise HTTPException(status_code=400, detail="Invalid token format")
+        elif "insufficient" in error_msg.lower() or "melt fee" in error_msg.lower():
+            raise HTTPException(
+                status_code=400,
+                detail=f"Token value is too small to cover swap fees. {error_msg}",
+            )
+        elif "failed to melt" in error_msg.lower():
+            raise HTTPException(
+                status_code=400,
+                detail=f"Failed to swap foreign mint token. {error_msg}",
+            )
         else:
-            raise HTTPException(status_code=400, detail="Failed to redeem token")
-    except Exception:
+            raise HTTPException(status_code=400, detail=f"Failed to redeem token: {error_msg}")
+    except Exception as e:
+        logger.error(
+            "topup_wallet_endpoint: unhandled error",
+            extra={"error": str(e), "error_type": type(e).__name__},
+        )
         raise HTTPException(status_code=500, detail="Internal server error")
     return {"msats": amount_msats}
 
@@ -154,30 +211,95 @@ async def _refund_cache_set(authorization: str, value: dict[str, str]) -> None:
         _refund_cache[key] = (expiry, value)
 
 
-@router.post("/refund")
+async def _restore_balance(
+    session: AsyncSession, hashed_key: str, balance: int, reserved_balance: int
+) -> None:
+    """Restore balance after a failed refund mint attempt."""
+    restore_stmt = (
+        update(ApiKey)
+        .where(col(ApiKey.hashed_key) == hashed_key)
+        .values(
+            balance=col(ApiKey.balance) + balance,
+            reserved_balance=col(ApiKey.reserved_balance) + reserved_balance,
+        )
+    )
+    await session.exec(restore_stmt)  # type: ignore[call-overload]
+    await session.commit()
+    logger.info(
+        "refund_wallet_endpoint: balance restored after mint failure",
+        extra={"hashed_key": hashed_key, "restored_balance": balance},
+    )
+
+
+@router.post("/refund", response_model=None)
 async def refund_wallet_endpoint(
-    authorization: Annotated[str, Header(...)],
+    authorization: Annotated[str | None, Header()] = None,
+    x_cashu: Annotated[str | None, Header()] = None,
     session: AsyncSession = Depends(get_session),
-) -> dict[str, str]:
-    if not authorization.startswith("Bearer "):
+) -> JSONResponse | dict[str, str]:
+    if x_cashu:
+        # Find the "in" transaction by the original payment token
+        in_tx_result = await session.exec(
+            select(CashuTransaction).where(
+                CashuTransaction.token == x_cashu,
+                CashuTransaction.type == "in",
+            )
+        )
+        in_tx = in_tx_result.first()
+        if in_tx is None:
+            raise HTTPException(status_code=404, detail="Refund not found")
+
+        # Use the request_id to find the associated "out" (refund) transaction
+        if in_tx.request_id is None:
+            raise HTTPException(status_code=404, detail="Refund not found")
+
+        out_tx_result = await session.exec(
+            select(CashuTransaction).where(
+                CashuTransaction.request_id == in_tx.request_id,
+                CashuTransaction.type == "out",
+            )
+        )
+        out_tx = out_tx_result.first()
+        if out_tx is None:
+            raise HTTPException(status_code=404, detail="Refund not found")
+        if out_tx.swept:
+            raise HTTPException(status_code=410, detail="Refund has been swept")
+
+        out_tx.collected = True
+        session.add(out_tx)
+        await session.commit()
+        body: dict[str, str] = {"token": out_tx.token}
+        if out_tx.unit == "sat":
+            body["sats"] = str(out_tx.amount)
+        else:
+            body["msats"] = str(out_tx.amount)
+        return JSONResponse(content=body, headers={"X-Cashu": out_tx.token})
+
+    if authorization is None or not authorization.startswith("Bearer "):
         raise HTTPException(
             status_code=401,
             detail="Invalid authorization. Use 'Bearer <cashu-token>' or 'Bearer <api-key>'",
         )
 
     bearer_value: str = authorization[7:]
-
-    if cached := await _refund_cache_get(bearer_value):
-        return cached
-
     key: ApiKey = await validate_bearer_key(bearer_value, session)
 
+    if key.total_balance <= 0:
+        if cached := await _refund_cache_get(bearer_value):
+            return cached
+
     if key.parent_key_hash:
         raise HTTPException(
             status_code=400,
             detail="Cannot refund child key. Please refund the parent key instead.",
         )
 
+    if key.reserved_balance > 0:
+        raise HTTPException(
+            status_code=400,
+            detail="Cannot refund key. There are ongoing requests for this api key.",
+        )
+
     remaining_balance_msats: int = key.total_balance
 
     if key.refund_currency == "sat":
@@ -190,7 +312,31 @@ async def refund_wallet_endpoint(
     elif remaining_balance <= 0:
         raise HTTPException(status_code=400, detail="No balance to refund")
 
-    # Perform refund operation first, before modifying balance
+    # Capture values before debit — the session may refresh key after commit
+    pre_debit_balance = key.balance
+    pre_debit_reserved = key.reserved_balance
+
+    # --- DEBIT FIRST: atomically zero the balance before minting tokens ---
+    # This prevents the race where a concurrent topup/spend happens between
+    # reading the balance and minting the refund token (double-spend).
+    debit_stmt = (
+        update(ApiKey)
+        .where(col(ApiKey.hashed_key) == key.hashed_key)
+        .where(col(ApiKey.balance) == pre_debit_balance)
+        .where(col(ApiKey.reserved_balance) == pre_debit_reserved)
+        .values(balance=0, reserved_balance=0)
+    )
+    debit_result = await session.exec(debit_stmt)  # type: ignore[call-overload]
+    await session.commit()
+
+    if debit_result.rowcount == 0:
+        # Balance changed between read and debit — another request is active
+        raise HTTPException(
+            status_code=409,
+            detail="Balance changed concurrently. Please retry the refund.",
+        )
+
+    # --- MINT: balance is locked at zero, safe to create the refund token ---
     try:
         if key.refund_address:
             from .core.settings import settings as global_settings
@@ -214,11 +360,24 @@ async def refund_wallet_endpoint(
         else:
             result["msats"] = str(remaining_balance_msats)
 
+        if "token" in result:
+            logger.info(
+                "refund_wallet_endpoint: cashu token issued",
+                extra={
+                    "path": "/v1/wallet/refund",
+                    "token": result["token"],
+                    "amount": remaining_balance,
+                    "currency": key.refund_currency or "sat",
+                },
+            )
+
     except HTTPException:
-        # Re-raise HTTP exceptions (like 400 for balance too small)
+        # Minting failed — restore the debited balance
+        await _restore_balance(session, key.hashed_key, pre_debit_balance, pre_debit_reserved)
         raise
     except Exception as e:
-        # If refund fails, don't modify the database
+        # Minting failed — restore the debited balance
+        await _restore_balance(session, key.hashed_key, pre_debit_balance, pre_debit_reserved)
         error_msg = str(e)
         if (
             "mint" in error_msg.lower()
@@ -232,12 +391,67 @@ async def refund_wallet_endpoint(
 
     await _refund_cache_set(bearer_value, result)
 
-    await session.delete(key)
-    await session.commit()
+    if "token" in result:
+        try:
+            await store_cashu_transaction(
+                token=result["token"],
+                amount=remaining_balance,
+                unit=key.refund_currency or "sat",
+                mint_url=key.refund_mint_url,
+                typ="out",
+                collected=False,
+                source="apikey",
+                api_key_hashed_key=key.hashed_key,
+            )
+        except Exception:
+            pass  # store_cashu_transaction already logs
+
+    logger.info(
+        "refund_wallet_endpoint: refund successful",
+        extra={
+            "refunded_msats": remaining_balance_msats,
+            "previous_reserved_balance": key.reserved_balance,
+        },
+    )
 
     return result
 
 
+@router.get("/history")
+async def wallet_history(
+    key: ApiKey = Depends(get_key_from_header),
+    session: AsyncSession = Depends(get_session),
+) -> dict[str, list[dict[str, str | int | bool | None]]]:
+    if key.parent_key_hash:
+        raise HTTPException(
+            status_code=400,
+            detail="Cannot view child key history. Please use the parent key instead.",
+        )
+
+    result = await session.exec(
+        select(CashuTransaction)
+        .where(CashuTransaction.api_key_hashed_key == key.hashed_key)
+        .order_by(col(CashuTransaction.created_at).desc())
+    )
+    transactions = result.all()
+    return {
+        "transactions": [
+            {
+                "id": tx.id,
+                "type": tx.type,
+                "source": tx.source,
+                "amount": tx.amount,
+                "unit": tx.unit,
+                "mint_url": tx.mint_url,
+                "created_at": tx.created_at,
+                "collected": tx.collected,
+                "swept": tx.swept,
+            }
+            for tx in transactions
+        ]
+    }
+
+
 @router.post("/donate")
 async def donate(token: str, ref: str | None = None) -> str:
     try:
@@ -253,6 +467,9 @@ async def donate(token: str, ref: str | None = None) -> str:
 
 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")
@@ -302,6 +519,12 @@ async def create_child_key(
             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)
@@ -320,6 +543,60 @@ async def create_child_key(
     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.get("/cashu-refund/{payment_token_hash}")
+async def get_cashu_refund(
+    payment_token_hash: str,
+    session: AsyncSession = Depends(get_session),
+) -> dict:
+    """Retrieve a stored Cashu refund token by the hash of the original payment token."""
+    result = await session.get(CashuTransaction, payment_token_hash)
+    if result is None:
+        raise HTTPException(status_code=404, detail="Refund not found")
+    if result.swept:
+        raise HTTPException(status_code=410, detail="Refund has been swept")
+    result.collected = True
+    session.add(result)
+    await session.commit()
+    return {
+        "refund_token": result.token,
+        "amount": result.amount,
+        "unit": result.unit,
+    }
+
+
 @router.api_route(
     "/{path:path}",
     methods=["GET", "POST", "PUT", "DELETE"],

routstr/core/db.py

@@ -1,13 +1,18 @@
 import os
+import pathlib
+import sqlite3
 import time
+import uuid
 from contextlib import asynccontextmanager
 from typing import AsyncGenerator
 
 from alembic import command
 from alembic.config import Config
+from alembic.util.exc import CommandError
 from sqlalchemy import UniqueConstraint
+from sqlalchemy.exc import OperationalError
 from sqlalchemy.ext.asyncio.engine import create_async_engine
-from sqlmodel import Field, Relationship, SQLModel, func, select, update
+from sqlmodel import Field, Relationship, SQLModel, col, func, select, update
 from sqlmodel.ext.asyncio.session import AsyncSession
 
 from .logging import get_logger
@@ -51,6 +56,22 @@ class ApiKey(SQLModel, table=True):  # type: ignore
     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:
@@ -85,6 +106,10 @@ class ModelRow(SQLModel, table=True):  # type: ignore
         default=None, description="JSON array of model alias IDs"
     )
     enabled: bool = Field(default=True, description="Whether this model is enabled")
+    forwarded_model_id: str | None = Field(
+        default=None,
+        description="Model ID to use when forwarding requests to upstream provider. Defaults to id if not set.",
+    )
     upstream_provider: "UpstreamProviderRow" = Relationship(back_populates="models")
 
 
@@ -110,10 +135,79 @@ class LightningInvoice(SQLModel, table=True):  # type: ignore
     paid_at: int | None = Field(default=None, description="Unix timestamp when paid")
 
 
+class CashuTransaction(SQLModel, table=True):  # type: ignore
+    __tablename__ = "cashu_transactions"
+
+    id: str = Field(
+        primary_key=True,
+        default_factory=lambda: uuid.uuid4().hex,
+        description="Unique transaction identifier",
+    )
+    token: str = Field(description="Serialized Cashu token")
+    amount: int = Field(description="Amount in the token's unit")
+    unit: str = Field(description="Token unit (sat or msat)")
+    mint_url: str | None = Field(default=None, description="Mint URL for the token")
+    type: str = Field(default="out", description="Transaction type: in or out")
+    request_id: str | None = Field(default=None, description="Associated request ID")
+    created_at: int = Field(
+        default_factory=lambda: int(time.time()),
+        description="Unix timestamp",
+    )
+    collected: bool = Field(default=False)
+    swept: bool = Field(default=False)
+    source: str = Field(
+        default="x-cashu",
+        description="Payment source: x-cashu or apikey",
+    )
+    api_key_hashed_key: str | None = Field(
+        default=None,
+        foreign_key="api_keys.hashed_key",
+        index=True,
+        description="Associated API key hash for wallet history",
+    )
+
+
+async def store_cashu_transaction(
+    token: str,
+    amount: int,
+    unit: str,
+    mint_url: str | None = None,
+    typ: str = "out",
+    request_id: str | None = None,
+    collected: bool = False,
+    created_at: int | None = None,
+    source: str = "x-cashu",
+    api_key_hashed_key: str | None = None,
+) -> None:
+    try:
+        async with create_session() as session:
+            tx = CashuTransaction(
+                token=token,
+                amount=amount,
+                unit=unit,
+                mint_url=mint_url,
+                type=typ,
+                request_id=request_id,
+                collected=collected,
+                created_at=created_at or int(time.time()),
+                source=source,
+                api_key_hashed_key=api_key_hashed_key,
+            )
+            session.add(tx)
+            await session.commit()
+    except Exception as e:
+        logger.warning(
+            f"Failed to store cashu transaction: {e} (type={typ})",
+            extra={"error": str(e), "type": typ},
+        )
+
+
 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"),
+        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(
@@ -126,7 +220,16 @@ class UpstreamProviderRow(SQLModel, table=True):  # type: ignore
     )
     enabled: bool = Field(default=True, description="Whether this provider is enabled")
     provider_fee: float = Field(
-        default=1.01, description="Provider fee multiplier (default 1%)"
+        default=1.01, description="Active fee multiplier (can be set by schedule)"
+    )
+    provider_fee_default: float = Field(
+        default=1.01, description="Default fee multiplier (outside schedules)"
+    )
+    provider_settings: str | None = Field(
+        default=None, description="JSON string for provider-specific settings"
+    )
+    provider_fee_schedules: str | None = Field(
+        default=None, description="JSON array of fee time ranges (HH:MM UTC)"
     )
     models: list["ModelRow"] = Relationship(
         back_populates="upstream_provider",
@@ -134,6 +237,66 @@ class UpstreamProviderRow(SQLModel, table=True):  # type: ignore
     )
 
 
+class RoutstrFee(SQLModel, table=True):  # type: ignore
+    __tablename__ = "routstr_fees"
+    id: int = Field(default=1, primary_key=True)
+    accumulated_msats: int = Field(default=0)
+    total_paid_msats: int = Field(default=0)
+    last_paid_at: int | None = Field(default=None)
+
+
+class CliToken(SQLModel, table=True):  # type: ignore
+    """Long-lived authorization token for CLI/agent use against admin endpoints."""
+
+    __tablename__ = "cli_tokens"
+    id: str = Field(
+        primary_key=True, default_factory=lambda: uuid.uuid4().hex
+    )
+    token: str = Field(unique=True, index=True, description="Bearer token value")
+    name: str = Field(description="Human-readable label for this token")
+    created_at: int = Field(default_factory=lambda: int(time.time()))
+    last_used_at: int | None = Field(default=None)
+    expires_at: int | None = Field(
+        default=None, description="Optional expiry unix timestamp; null = never expires"
+    )
+
+
+async def accumulate_routstr_fee(session: AsyncSession, amount_msats: int) -> None:
+    stmt = (
+        update(RoutstrFee)
+        .where(col(RoutstrFee.id) == 1)
+        .values(accumulated_msats=RoutstrFee.accumulated_msats + amount_msats)
+    )
+    result = await session.exec(stmt)  # type: ignore[call-overload]
+    if result.rowcount == 0:
+        session.add(RoutstrFee(id=1, accumulated_msats=amount_msats))
+    await session.commit()
+
+
+async def get_routstr_fee(session: AsyncSession) -> RoutstrFee:
+    fee = await session.get(RoutstrFee, 1)
+    if fee is None:
+        fee = RoutstrFee(id=1, accumulated_msats=0, total_paid_msats=0)
+        session.add(fee)
+        await session.commit()
+        await session.refresh(fee)
+    return fee
+
+
+async def reset_routstr_fee(session: AsyncSession, paid_msats: int) -> None:
+    stmt = (
+        update(RoutstrFee)
+        .where(col(RoutstrFee.id) == 1)
+        .values(
+            accumulated_msats=RoutstrFee.accumulated_msats - paid_msats,
+            total_paid_msats=RoutstrFee.total_paid_msats + paid_msats,
+            last_paid_at=int(time.time()),
+        )
+    )
+    await session.exec(stmt)  # type: ignore[call-overload]
+    await session.commit()
+
+
 async def balances_for_mint_and_unit(
     db_session: AsyncSession, mint_url: str, unit: str
 ) -> int:
@@ -163,11 +326,64 @@ async def create_session() -> AsyncGenerator[AsyncSession, None]:
         yield session
 
 
+def fix_cashu_migrations() -> None:
+    """
+    Fixes Cashu wallet migrations that are not idempotent.
+    This specifically addresses the 'duplicate column name: public_keys' error
+    in the keysets table of Cashu's internal SQLite databases.
+    """
+    project_root = pathlib.Path(__file__).resolve().parents[2]
+    wallet_dir = project_root / ".wallet"
+
+    if not wallet_dir.exists() or not wallet_dir.is_dir():
+        return
+
+    logger.info("Checking Cashu wallet databases for migration idempotency")
+
+    for db_file in wallet_dir.glob("*.sqlite3"):
+        try:
+            conn = sqlite3.connect(db_file)
+            cursor = conn.cursor()
+
+            # Check if keysets table exists
+            cursor.execute(
+                "SELECT name FROM sqlite_master WHERE type='table' AND name='keysets'"
+            )
+            if not cursor.fetchone():
+                conn.close()
+                continue
+
+            # Check if public_keys column exists
+            cursor.execute("PRAGMA table_info(keysets)")
+            columns = [info[1] for info in cursor.fetchall()]
+
+            if "public_keys" not in columns:
+                logger.info(f"Adding missing public_keys column to {db_file.name}")
+                cursor.execute("ALTER TABLE keysets ADD COLUMN public_keys TEXT")
+                conn.commit()
+
+            conn.close()
+        except Exception as e:
+            logger.warning(f"Could not check/fix Cashu database {db_file}: {e}")
+
+
+def _clear_alembic_version() -> None:
+    """Clear the alembic_version table so stamp/upgrade can proceed."""
+    sync_url = DATABASE_URL.replace("+aiosqlite", "")
+    from sqlalchemy import create_engine, text
+
+    eng = create_engine(sync_url)
+    with eng.begin() as conn:
+        conn.execute(text("DELETE FROM alembic_version"))
+    eng.dispose()
+
+
 def run_migrations() -> None:
     """Run Alembic migrations programmatically."""
-    import pathlib
-
     try:
+        # Run Cashu migration fix first
+        fix_cashu_migrations()
+
         # Get the path to the alembic.ini file
         project_root = pathlib.Path(__file__).resolve().parents[2]
         alembic_ini_path = project_root / "alembic.ini"
@@ -183,8 +399,30 @@ def run_migrations() -> None:
         # Set the database URL in the config
         alembic_cfg.set_main_option("sqlalchemy.url", DATABASE_URL)
 
-        # Run migrations to the latest revision
-        command.upgrade(alembic_cfg, "head")
+        try:
+            command.upgrade(alembic_cfg, "head")
+        except CommandError as e:
+            if "Can't locate revision" in str(e):
+                logger.warning(
+                    "Database stamped with unknown revision (likely from another branch). "
+                    "Re-stamping to current head.",
+                    extra={"error": str(e)},
+                )
+                _clear_alembic_version()
+                command.stamp(alembic_cfg, "head")
+            else:
+                raise
+        except OperationalError as e:
+            if "duplicate column name" in str(e).lower():
+                logger.warning(
+                    "Migration hit a column that already exists (likely added via "
+                    "create_all on another branch). Stamping to current head.",
+                    extra={"error": str(e)},
+                )
+                _clear_alembic_version()
+                command.stamp(alembic_cfg, "head")
+            else:
+                raise
 
         logger.info("Database migrations completed successfully")
 

routstr/core/logging.py

@@ -3,36 +3,38 @@ Logging configuration for Routstr.
 
 CRITICAL LOG MESSAGES FOR USAGE STATISTICS:
 ===========================================
-The following log messages are parsed by the usage tracking system (routstr/core/admin.py).
+The following log messages are parsed by the usage tracking system
+(routstr/core/usage_analytics_store.py and routstr/core/log_manager.py).
 DO NOT modify or remove these messages without updating the usage tracking logic:
 
 1. "Received proxy request" (INFO) - routstr/proxy.py
    - Used to count total incoming requests
    - Includes model information in context
 
-    2. "Payment adjustment completed for streaming" (INFO) - routstr/upstream/base.py
-       "Payment adjustment completed for non-streaming" (INFO) - routstr/upstream/base.py
+2. "Calculated token-based cost" (INFO) - routstr/auth.py
    - Used to track successful completions and revenue
-   - The 'cost_data.total_msats' field is extracted for revenue calculation
-   - Must include 'cost_data' in extra dict
+   - The 'token_cost', 'model', 'input_tokens', and 'output_tokens' fields are extracted for dashboard metrics
 
-3. "Payment processed successfully" (INFO) - routstr/auth.py
+3. "Max cost payment finalized" (INFO) - routstr/auth.py
+   - Used as the successful completion fallback when token usage is unavailable
+   - The 'charged_amount', 'model', 'input_tokens', and 'output_tokens' fields are extracted for dashboard metrics
+
+4. "Payment processed successfully" (INFO) - routstr/auth.py
    - Used to count successful payment processing events
    - Tracks payment-related metrics
 
-4. "Upstream request failed, revert payment" (WARNING) - routstr/proxy.py
+5. "Upstream request failed, revert payment" (WARNING) - routstr/proxy.py
    - Used to track failed requests and refunds
    - The 'max_cost_for_model' field is extracted for refund calculation
    - Must include 'max_cost_for_model' in extra dict
 
-5. Any ERROR level logs with "upstream" in the message
+6. Any ERROR level logs with "upstream" in the message
    - Used to count upstream provider errors
    - Helps identify service reliability issues
 
 If you need to modify these messages, ensure you also update the parsing logic in:
-- routstr/core/admin.py:_aggregate_metrics_by_time()
-- routstr/core/admin.py:_get_summary_stats()
-- routstr/core/admin.py:get_revenue_by_model()
+- routstr/core/usage_analytics_store.py
+- routstr/core/log_manager.py
 """
 
 import logging.config

routstr/core/main.py

@@ -10,13 +10,19 @@ 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 ..nostr import (
+    announce_provider,
+    providers_cache_refresher,
+    publish_usage_analytics,
+)
+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
+from ..upstream.auto_topup import periodic_auto_topup
+from ..wallet import periodic_payout, periodic_refund_sweep, periodic_routstr_fee_payout
 from .admin import admin_router
 from .db import create_session, init_db, run_migrations
 from .exceptions import general_exception_handler, http_exception_handler
@@ -30,9 +36,9 @@ setup_logging()
 logger = get_logger(__name__)
 
 if os.getenv("VERSION_SUFFIX") is not None:
-    __version__ = f"0.3.0-{os.getenv('VERSION_SUFFIX')}"
+    __version__ = f"0.4.3-{os.getenv('VERSION_SUFFIX')}"
 else:
-    __version__ = "0.3.0"
+    __version__ = "0.4.3"
 
 
 @asynccontextmanager
@@ -43,9 +49,14 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
     pricing_task = None
     payout_task = None
     nip91_task = None
+    analytics_task = None
     providers_task = None
     models_refresh_task = None
     model_maps_refresh_task = None
+    key_reset_task = None
+    auto_topup_task = None
+    refund_sweep_task = None
+    routstr_fee_task = None
 
     try:
         # Run database migrations on startup
@@ -92,15 +103,23 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
         btc_price_task = asyncio.create_task(update_prices_periodically())
         pricing_task = asyncio.create_task(update_sats_pricing())
         if global_settings.models_refresh_interval_seconds > 0:
+            # Pass the accessor (not its current value) so the loop sees providers
+            # added/changed via reinitialize_upstreams() instead of staying pinned
+            # to the startup snapshot.
             models_refresh_task = asyncio.create_task(
-                refresh_upstreams_models_periodically(get_upstreams())
+                refresh_upstreams_models_periodically(get_upstreams)
             )
         model_maps_refresh_task = asyncio.create_task(refresh_model_maps_periodically())
         payout_task = asyncio.create_task(periodic_payout())
         if global_settings.nsec:
             nip91_task = asyncio.create_task(announce_provider())
+        analytics_task = asyncio.create_task(publish_usage_analytics())
         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())
+        auto_topup_task = asyncio.create_task(periodic_auto_topup())
+        refund_sweep_task = asyncio.create_task(periodic_refund_sweep())
+        routstr_fee_task = asyncio.create_task(periodic_routstr_fee_payout())
 
         yield
 
@@ -124,12 +143,22 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
             payout_task.cancel()
         if nip91_task is not None:
             nip91_task.cancel()
+        if analytics_task is not None:
+            analytics_task.cancel()
         if providers_task is not None:
             providers_task.cancel()
         if models_refresh_task is not 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()
+        if auto_topup_task is not None:
+            auto_topup_task.cancel()
+        if refund_sweep_task is not None:
+            refund_sweep_task.cancel()
+        if routstr_fee_task is not None:
+            routstr_fee_task.cancel()
 
         try:
             tasks_to_wait = []
@@ -141,12 +170,22 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
                 tasks_to_wait.append(payout_task)
             if nip91_task is not None:
                 tasks_to_wait.append(nip91_task)
+            if analytics_task is not None:
+                tasks_to_wait.append(analytics_task)
             if providers_task is not None:
                 tasks_to_wait.append(providers_task)
             if models_refresh_task is not 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 auto_topup_task is not None:
+                tasks_to_wait.append(auto_topup_task)
+            if refund_sweep_task is not None:
+                tasks_to_wait.append(refund_sweep_task)
+            if routstr_fee_task is not None:
+                tasks_to_wait.append(routstr_fee_task)
 
             if tasks_to_wait:
                 await asyncio.gather(*tasks_to_wait, return_exceptions=True)
@@ -167,7 +206,7 @@ app.add_middleware(
     allow_credentials=True,
     allow_methods=["*"],
     allow_headers=["*"],
-    expose_headers=["x-routstr-request-id"],
+    expose_headers=["x-routstr-request-id", "x-cashu"],
 )
 
 # Add logging middleware

routstr/core/middleware.py

@@ -38,11 +38,6 @@ class LoggingMiddleware(BaseHTTPMiddleware):
             except Exception:
                 pass
 
-        # Extract request info
-        client_host = None
-        if request.client:
-            client_host = request.client.host
-
         # Log incoming request
         logger.info(
             "Incoming request",
@@ -51,7 +46,6 @@ class LoggingMiddleware(BaseHTTPMiddleware):
                 "method": request.method,
                 "path": request.url.path,
                 "query_params": dict(request.query_params),
-                "client_host": client_host,
                 "headers": {
                     k: v
                     for k, v in request.headers.items()
@@ -100,7 +94,6 @@ class LoggingMiddleware(BaseHTTPMiddleware):
                     "path": request.url.path,
                     "status_code": response.status_code,
                     "duration_ms": round(duration * 1000, 2),
-                    "client_host": client_host,
                 },
             )
             if hasattr(response, "headers"):
@@ -120,7 +113,6 @@ class LoggingMiddleware(BaseHTTPMiddleware):
                     "method": request.method,
                     "path": request.url.path,
                     "duration_ms": round(duration * 1000, 2),
-                    "client_host": client_host,
                     "error": str(e),
                     "error_type": type(e).__name__,
                 },

routstr/core/settings.py

@@ -52,7 +52,7 @@ 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")
+    child_key_cost: int = Field(default=0, 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(
@@ -74,6 +74,7 @@ class Settings(BaseSettings):
     enable_pricing_refresh: bool = Field(default=True, env="ENABLE_PRICING_REFRESH")
     enable_models_refresh: bool = Field(default=True, env="ENABLE_MODELS_REFRESH")
     refund_cache_ttl_seconds: int = Field(default=3600, env="REFUND_CACHE_TTL_SECONDS")
+    refund_sweep_ttl_seconds: int = Field(default=604800, env="REFUND_SWEEP_TTL_SECONDS")
 
     # Logging
     log_level: str = Field(default="INFO", env="LOG_LEVEL")
@@ -92,6 +93,20 @@ class Settings(BaseSettings):
 
     # Discovery
     relays: list[str] = Field(default_factory=list, env="RELAYS")
+    enable_analytics_sharing: bool = Field(
+        default=True, env="ENABLE_ANALYTICS_SHARING"
+    )
+
+def _normalize_settings_data(data: dict[str, Any]) -> dict[str, Any]:
+    """Discard unknown keys from persisted settings."""
+    normalized: dict[str, Any] = {}
+    known_fields = Settings.__fields__
+
+    for key, value in data.items():
+        if key in known_fields:
+            normalized[key] = value
+
+    return normalized
 
 
 def _compute_primary_mint(cashu_mints: list[str]) -> str:
@@ -143,7 +158,7 @@ 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:
@@ -231,16 +246,21 @@ class SettingsService:
 
             db_id, db_data, _updated_at = row
             try:
-                db_json = (
+                db_json_raw = (
                     json.loads(db_data) if isinstance(db_data, str) else dict(db_data)
                 )
+                if not isinstance(db_json_raw, dict):
+                    db_json_raw = {}
             except Exception:
-                db_json = {}
+                db_json_raw = {}
+            db_json = _normalize_settings_data(db_json_raw)
 
+            valid_fields = set(env_resolved.dict().keys())
             merged_dict: dict[str, Any] = dict(env_resolved.dict())
             merged_dict.update(
-                {k: v for k, v in db_json.items() if v not in (None, "", [], {})}
+                {k: v for k, v in db_json.items() if v not in (None, "", [], {}) and k in valid_fields}
             )
+            merged_dict = Settings(**merged_dict).dict()
 
             # Ensure primary_mint is consistent with cashu_mints if not explicitly set
             if not merged_dict.get("primary_mint"):
@@ -248,7 +268,7 @@ class SettingsService:
                     merged_dict.get("cashu_mints", [])
                 )
 
-            if any(k not in db_json for k in merged_dict.keys()):
+            if db_json_raw != merged_dict:
                 await db_session.exec(  # type: ignore
                     text(
                         "UPDATE settings SET data = :data, updated_at = :updated_at WHERE id = 1"
@@ -271,7 +291,7 @@ class SettingsService:
     ) -> Settings:
         async with cls._lock:
             current = cls.get()
-            candidate_dict = {**current.dict(), **partial}
+            candidate_dict = {**current.dict(), **_normalize_settings_data(partial)}
             candidate = Settings(**candidate_dict)
             from sqlmodel import text
 
@@ -305,8 +325,10 @@ class SettingsService:
                 raise RuntimeError("Settings row missing")
             (data_str,) = row
             data = json.loads(data_str) if isinstance(data_str, str) else dict(data_str)
+            valid_fields = set(settings.dict().keys())
             # Update in-place
             for k, v in data.items():
-                setattr(settings, k, v)
+                if k in valid_fields:
+                    setattr(settings, k, v)
             cls._current = settings
             return settings

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 +1,5 @@
+from .analytics import publish_usage_analytics
+from .discovery import providers_cache_refresher
+from .listing import announce_provider
+
+__all__ = ["providers_cache_refresher", "announce_provider", "publish_usage_analytics"]

routstr/nostr/analytics.py

@@ -0,0 +1,419 @@
+#!/usr/bin/env python3
+"""
+Nostr usage analytics publisher.
+Publishes a single replaceable analytics snapshot for each provider.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import hashlib
+import json
+import time
+from typing import Any
+
+from nostr.event import Event
+from nostr.key import PrivateKey
+
+from ..core import get_logger
+from ..core.log_manager import log_manager
+from ..core.settings import settings
+from .listing import nsec_to_keypair, publish_to_relay
+
+logger = get_logger(__name__)
+
+ANALYTICS_KIND = 38422
+ANALYTICS_SCHEMA = "routstr.analytics.snapshot.v1"
+DEFAULT_RELAYS = [
+    "wss://relay.nostr.band",
+    "wss://relay.damus.io",
+    "wss://relay.routstr.com",
+    "wss://nos.lol",
+]
+PUBLISH_INTERVAL_SECONDS = 15 * 60
+DISABLED_POLL_SECONDS = 60
+DASHBOARD_WINDOW_HOURS = 24
+DASHBOARD_INTERVAL_MINUTES = 60
+MODEL_LIMIT = 20
+WINDOW_DEFINITIONS: tuple[tuple[str, int, int], ...] = (
+    ("24h", 24, 60),
+    ("7d", 7 * 24, 6 * 60),
+    ("30d", 30 * 24, 24 * 60),
+    ("3m", 90 * 24, 24 * 60),
+    ("1y", 365 * 24, 7 * 24 * 60),
+)
+
+
+def _event_to_dict(ev: Event) -> dict[str, Any]:
+    return {
+        "id": ev.id,
+        "pubkey": ev.public_key,
+        "created_at": ev.created_at,
+        "kind": int(ev.kind) if not isinstance(ev.kind, int) else ev.kind,
+        "tags": ev.tags,
+        "content": ev.content,
+        "sig": ev.signature,
+    }
+
+
+def _resolve_provider_id(public_key_hex: str) -> str:
+    explicit_provider_id = (settings.provider_id or "").strip()
+    if explicit_provider_id:
+        return explicit_provider_id
+    return public_key_hex[:12]
+
+
+def _resolve_endpoint_urls() -> list[str]:
+    urls: list[str] = []
+    http_url = (settings.http_url or "").strip()
+    onion_url = (settings.onion_url or "").strip()
+
+    if http_url and http_url != "http://localhost:8000":
+        urls.append(http_url)
+
+    if onion_url:
+        if onion_url.endswith(".onion") and not (
+            onion_url.startswith("http://") or onion_url.startswith("https://")
+        ):
+            onion_url = f"http://{onion_url}"
+        urls.append(onion_url)
+
+    return urls
+
+
+def _resolve_relays() -> list[str]:
+    configured = [url.strip() for url in settings.relays if url.strip()]
+    return configured if configured else list(DEFAULT_RELAYS)
+
+
+def _to_int(value: Any) -> int:
+    if isinstance(value, bool):
+        return int(value)
+    if isinstance(value, int):
+        return value
+    if isinstance(value, float):
+        return int(value)
+    if isinstance(value, str):
+        try:
+            return int(float(value))
+        except ValueError:
+            return 0
+    return 0
+
+
+def _to_float(value: Any) -> float:
+    if isinstance(value, bool):
+        return float(int(value))
+    if isinstance(value, (int, float)):
+        return float(value)
+    if isinstance(value, str):
+        try:
+            return float(value)
+        except ValueError:
+            return 0.0
+    return 0.0
+
+
+def _aggregate_top_model_usage(
+    model_usage_mix: dict[str, Any],
+) -> tuple[list[dict[str, Any]], dict[str, Any]]:
+    top_models_raw = model_usage_mix.get("top_models", [])
+    mix_metrics_raw = model_usage_mix.get("metrics", [])
+
+    top_models = [model for model in top_models_raw if isinstance(model, str)]
+    metrics = [row for row in mix_metrics_raw if isinstance(row, dict)]
+
+    model_totals: dict[str, dict[str, float | int]] = {
+        model: {
+            "successful_requests": 0,
+            "revenue_msats": 0.0,
+            "total_tokens": 0,
+        }
+        for model in top_models
+    }
+    others = {
+        "successful_requests": 0,
+        "revenue_msats": 0.0,
+        "total_tokens": 0,
+    }
+
+    for metric in metrics:
+        model_counts = metric.get("model_counts", {})
+        model_revenue = metric.get("model_revenue_msats", {})
+        model_tokens = metric.get("model_tokens", {})
+
+        if isinstance(model_counts, dict):
+            for model, count in model_counts.items():
+                if model in model_totals:
+                    model_totals[model]["successful_requests"] += _to_int(count)
+
+        if isinstance(model_revenue, dict):
+            for model, amount in model_revenue.items():
+                if model in model_totals:
+                    model_totals[model]["revenue_msats"] += _to_float(amount)
+
+        if isinstance(model_tokens, dict):
+            for model, token_count in model_tokens.items():
+                if model in model_totals:
+                    model_totals[model]["total_tokens"] += _to_int(token_count)
+
+        others["successful_requests"] += _to_int(metric.get("others", 0))
+        others["revenue_msats"] += _to_float(metric.get("others_revenue_msats", 0.0))
+        others["total_tokens"] += _to_int(metric.get("others_tokens", 0))
+
+    model_rows = [
+        {
+            "model": model,
+            "successful_requests": int(values["successful_requests"]),
+            "revenue_msats": float(values["revenue_msats"]),
+            "total_tokens": int(values["total_tokens"]),
+        }
+        for model, values in model_totals.items()
+    ]
+    model_rows.sort(
+        key=lambda row: _to_int(row.get("successful_requests", 0)),
+        reverse=True,
+    )
+
+    return model_rows, others
+
+
+def _build_summary_payload(summary: dict[str, Any]) -> dict[str, Any]:
+    return {
+        "total_requests": _to_int(summary.get("total_requests", 0)),
+        "successful_chat_completions": _to_int(
+            summary.get("successful_chat_completions", 0)
+        ),
+        "failed_requests": _to_int(summary.get("failed_requests", 0)),
+        "success_rate": _to_float(summary.get("success_rate", 0.0)),
+        "unique_models_count": _to_int(summary.get("unique_models_count", 0)),
+        "input_tokens": _to_int(summary.get("input_tokens", 0)),
+        "output_tokens": _to_int(summary.get("output_tokens", 0)),
+        "total_tokens": _to_int(summary.get("total_tokens", 0)),
+        "revenue_msats": _to_float(summary.get("revenue_msats", 0.0)),
+        "refunds_msats": _to_float(summary.get("refunds_msats", 0.0)),
+        "net_revenue_msats": _to_float(summary.get("net_revenue_msats", 0.0)),
+        "revenue_sats": _to_float(summary.get("revenue_sats", 0.0)),
+        "refunds_sats": _to_float(summary.get("refunds_sats", 0.0)),
+        "net_revenue_sats": _to_float(summary.get("net_revenue_sats", 0.0)),
+    }
+
+
+def _build_window_payload(
+    *,
+    hours: int,
+    interval_minutes: int,
+    model_limit: int,
+) -> dict[str, Any]:
+    dashboard = log_manager.get_usage_dashboard(
+        interval=interval_minutes,
+        hours=hours,
+        error_limit=1,
+        model_limit=model_limit,
+    )
+
+    summary = dashboard.get("summary", {})
+    model_usage_mix = dashboard.get("model_usage_mix", {})
+
+    summary_payload = _build_summary_payload(summary if isinstance(summary, dict) else {})
+    usage_mix_payload = model_usage_mix if isinstance(model_usage_mix, dict) else {}
+    top_model_usage, others_usage = _aggregate_top_model_usage(usage_mix_payload)
+
+    return {
+        "window_hours": hours,
+        "interval_minutes": interval_minutes,
+        "summary": summary_payload,
+        "model_usage_mix": usage_mix_payload,
+        "top_model_usage": top_model_usage,
+        "others_usage": others_usage,
+    }
+
+
+def build_stats_snapshot_payload(
+    provider_id: str,
+    *,
+    public_key_hex: str,
+    generated_at: int,
+    window_hours: int = DASHBOARD_WINDOW_HOURS,
+    interval_minutes: int = DASHBOARD_INTERVAL_MINUTES,
+    model_limit: int = MODEL_LIMIT,
+) -> dict[str, Any]:
+    _ = (window_hours, interval_minutes)
+    windows: dict[str, dict[str, Any]] = {}
+    for key, hours, window_interval_minutes in WINDOW_DEFINITIONS:
+        windows[key] = _build_window_payload(
+            hours=hours,
+            interval_minutes=window_interval_minutes,
+            model_limit=model_limit,
+        )
+
+    primary_window = windows.get("24h", {})
+    summary_payload = (
+        primary_window.get("summary", {})
+        if isinstance(primary_window.get("summary", {}), dict)
+        else {}
+    )
+    usage_mix_payload = (
+        primary_window.get("model_usage_mix", {})
+        if isinstance(primary_window.get("model_usage_mix", {}), dict)
+        else {}
+    )
+    top_model_usage = (
+        primary_window.get("top_model_usage", [])
+        if isinstance(primary_window.get("top_model_usage", []), list)
+        else []
+    )
+    others_usage = (
+        primary_window.get("others_usage", {})
+        if isinstance(primary_window.get("others_usage", {}), dict)
+        else {}
+    )
+
+    return {
+        "schema": ANALYTICS_SCHEMA,
+        "generated_at": generated_at,
+        "provider_id": provider_id,
+        "pubkey": public_key_hex,
+        "npub": settings.npub or "",
+        "endpoint_urls": _resolve_endpoint_urls(),
+        "window_hours": DASHBOARD_WINDOW_HOURS,
+        "interval_minutes": DASHBOARD_INTERVAL_MINUTES,
+        "summary": summary_payload,
+        "model_usage_mix": usage_mix_payload,
+        "top_model_usage": top_model_usage,
+        "others_usage": others_usage,
+        "windows": windows,
+    }
+
+
+def create_stats_snapshot_event(
+    private_key_hex: str,
+    provider_id: str,
+    payload_json: str,
+    *,
+    d_tag: str,
+) -> dict[str, Any]:
+    private_key = PrivateKey(bytes.fromhex(private_key_hex))
+    tags = [
+        ["d", d_tag],
+        ["provider", provider_id],
+        ["schema", ANALYTICS_SCHEMA],
+    ]
+
+    event = Event(
+        public_key=private_key.public_key.hex(),
+        content=payload_json,
+        kind=ANALYTICS_KIND,
+        tags=tags,
+    )
+    private_key.sign_event(event)
+    return _event_to_dict(event)
+
+
+def _fingerprint_payload(payload: dict[str, Any]) -> str:
+    normalized = dict(payload)
+    # Ignore generated timestamp for semantic dedupe.
+    normalized.pop("generated_at", None)
+    payload_json = json.dumps(normalized, separators=(",", ":"), sort_keys=True)
+    return hashlib.sha256(payload_json.encode("utf-8")).hexdigest()
+
+
+async def publish_usage_analytics() -> None:
+    last_payload_hash: str | None = None
+
+    parsed_nsec: str | None = None
+    private_key_hex: str | None = None
+    public_key_hex: str | None = None
+    provider_id: str | None = None
+    warned_missing_nsec = False
+
+    logger.info("Usage analytics sharing task started")
+
+    while True:
+        try:
+            if not settings.enable_analytics_sharing:
+                await asyncio.sleep(DISABLED_POLL_SECONDS)
+                continue
+
+            nsec = (settings.nsec or "").strip()
+            if not nsec:
+                if not warned_missing_nsec:
+                    logger.info("NSEC is not configured; skipping analytics sharing to Nostr")
+                    warned_missing_nsec = True
+                await asyncio.sleep(DISABLED_POLL_SECONDS)
+                continue
+
+            warned_missing_nsec = False
+            if nsec != parsed_nsec or private_key_hex is None or public_key_hex is None:
+                keypair = nsec_to_keypair(nsec)
+                if not keypair:
+                    logger.error("Invalid NSEC; analytics sharing is paused")
+                    await asyncio.sleep(DISABLED_POLL_SECONDS)
+                    continue
+                private_key_hex, public_key_hex = keypair
+                parsed_nsec = nsec
+                provider_id = _resolve_provider_id(public_key_hex)
+                last_payload_hash = None
+
+            if private_key_hex is None or public_key_hex is None:
+                await asyncio.sleep(DISABLED_POLL_SECONDS)
+                continue
+
+            relay_urls = _resolve_relays()
+            if not relay_urls:
+                logger.warning("No Nostr relays configured; analytics sharing skipped")
+                await asyncio.sleep(DISABLED_POLL_SECONDS)
+                continue
+
+            resolved_provider_id = provider_id or _resolve_provider_id(public_key_hex)
+            now_ts = int(time.time())
+            payload = build_stats_snapshot_payload(
+                resolved_provider_id,
+                public_key_hex=public_key_hex,
+                generated_at=now_ts,
+            )
+
+            payload_hash = _fingerprint_payload(payload)
+            if last_payload_hash == payload_hash:
+                await asyncio.sleep(PUBLISH_INTERVAL_SECONDS)
+                continue
+
+            payload_json = json.dumps(payload, separators=(",", ":"), sort_keys=True)
+            d_tag = f"{resolved_provider_id}:stats"
+            event = create_stats_snapshot_event(
+                private_key_hex,
+                resolved_provider_id,
+                payload_json,
+                d_tag=d_tag,
+            )
+
+            success_count = 0
+            for relay_url in relay_urls:
+                if await publish_to_relay(relay_url, event):
+                    success_count += 1
+
+            if success_count > 0:
+                last_payload_hash = payload_hash
+
+            logger.info(
+                "Published analytics snapshot (success=%s/%s provider=%s)",
+                success_count,
+                len(relay_urls),
+                resolved_provider_id,
+                extra={
+                    "relay_success_count": success_count,
+                    "relay_total": len(relay_urls),
+                    "provider_id": resolved_provider_id,
+                },
+            )
+            await asyncio.sleep(PUBLISH_INTERVAL_SECONDS)
+
+        except asyncio.CancelledError:
+            logger.info("Usage analytics sharing task cancelled")
+            break
+        except Exception as e:
+            logger.error(
+                "Usage analytics sharing error",
+                extra={"error": str(e), "error_type": type(e).__name__},
+            )
+            await asyncio.sleep(DISABLED_POLL_SECONDS)

routstr/nostr/discovery.py

@@ -8,8 +8,8 @@ import httpx
 import websockets
 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")

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

@@ -16,6 +16,8 @@ class CostData(BaseModel):
     output_msats: int
     total_msats: int
     total_usd: float = 0.0
+    input_tokens: int = 0
+    output_tokens: int = 0
 
 
 class MaxCostData(CostData):
@@ -63,17 +65,64 @@ async def calculate_cost(  # todo: can be sync
             output_msats=0,
             total_msats=0,
             total_usd=0.0,
+            input_tokens=0,
+            output_tokens=0,
         )
 
     usage_data = response_data["usage"]
 
-    usd_cost = 0.0
+    def parse_token_count(value: object) -> int:
+        if isinstance(value, bool):
+            return 0
+        if isinstance(value, int):
+            return max(0, value)
+        if isinstance(value, float):
+            return max(0, int(value))
+        if isinstance(value, str):
+            try:
+                return max(0, int(float(value)))
+            except ValueError:
+                return 0
+        return 0
+
+    input_tokens = parse_token_count(usage_data.get("prompt_tokens", 0))
+    output_tokens = parse_token_count(usage_data.get("completion_tokens", 0))
+    input_tokens = (
+        input_tokens
+        if input_tokens != 0
+        else parse_token_count(usage_data.get("input_tokens", 0))
+    )
+    output_tokens = (
+        output_tokens
+        if output_tokens != 0
+        else parse_token_count(usage_data.get("output_tokens", 0))
+    )
+    input_tokens = (
+        input_tokens
+        if input_tokens != 0
+        else parse_token_count(response_data.get("usage", {}).get("input_tokens", 0))
+    )
+    output_tokens = (
+        output_tokens
+        if output_tokens != 0
+        else parse_token_count(response_data.get("usage", {}).get("output_tokens", 0))
+    )
+
+    usd_cost = 0.0
+    input_usd = 0.0
+    output_usd = 0.0
 
-    # Prioritize cost_details.upstream_inference_cost
     if "cost_details" in usage_data:
         usd_cost = float(
             usage_data["cost_details"].get("upstream_inference_cost", 0) or 0
         )
+        input_usd = float(
+            usage_data["cost_details"].get("upstream_inference_prompt_cost", 0) or 0
+        )
+        output_usd = float(
+            usage_data["cost_details"].get("upstream_inference_completions_cost", 0)
+            or 0
+        )
 
     # Fallback to cost field if upstream_inference_cost is 0
     if usd_cost == 0 and "cost" in usage_data:
@@ -82,12 +131,34 @@ async def calculate_cost(  # todo: can be sync
         except Exception:
             pass
 
+    MSATS_PER_1K_INPUT_TOKENS: float = (
+        float(settings.fixed_per_1k_input_tokens) * 1000.0
+    )
+    MSATS_PER_1K_OUTPUT_TOKENS: float = (
+        float(settings.fixed_per_1k_output_tokens) * 1000.0
+    )
+
     if usd_cost > 0:
         try:
             sats_per_usd = 1.0 / sats_usd_price()
             cost_in_sats = usd_cost * sats_per_usd
             cost_in_msats = math.ceil(cost_in_sats * 1000)
 
+            input_msats = 0
+            output_msats = 0
+
+            if input_usd > 0 or output_usd > 0:
+                input_msats = int((input_usd * sats_per_usd) * 1000)
+                output_msats = int((output_usd * sats_per_usd) * 1000)
+            else:
+                total_tokens = input_tokens + output_tokens
+                if total_tokens > 0:
+                    input_ratio = input_tokens / total_tokens
+                    input_msats = int(cost_in_msats * input_ratio)
+                    output_msats = cost_in_msats - input_msats
+                else:
+                    output_msats = cost_in_msats
+
             logger.info(
                 "Using cost from usage data/details",
                 extra={
@@ -99,11 +170,13 @@ async def calculate_cost(  # todo: can be sync
             )
 
             return CostData(
-                base_msats=-1,
-                input_msats=-1,  # Cost field doesn't break down by token type
-                output_msats=-1,
+                base_msats=0,
+                input_msats=input_msats,
+                output_msats=output_msats,
                 total_msats=cost_in_msats,
                 total_usd=usd_cost,
+                input_tokens=input_tokens,
+                output_tokens=output_tokens,
             )
         except Exception as e:
             logger.warning(
@@ -116,13 +189,6 @@ async def calculate_cost(  # todo: can be sync
             )
             # Fall through to token-based calculation
 
-    MSATS_PER_1K_INPUT_TOKENS: float = (
-        float(settings.fixed_per_1k_input_tokens) * 1000.0
-    )
-    MSATS_PER_1K_OUTPUT_TOKENS: float = (
-        float(settings.fixed_per_1k_output_tokens) * 1000.0
-    )
-
     if not settings.fixed_pricing:
         response_model = response_data.get("model", "")
         logger.debug(
@@ -184,35 +250,14 @@ async def calculate_cost(  # todo: can be sync
             input_msats=0,
             output_msats=0,
             total_msats=max_cost,
+            input_tokens=input_tokens,
+            output_tokens=output_tokens,
         )
 
-    input_tokens = usage_data.get("prompt_tokens", 0)
-    output_tokens = usage_data.get("completion_tokens", 0)
+    calc_input_msats = round(input_tokens / 1000 * MSATS_PER_1K_INPUT_TOKENS, 3)
 
-    # added for response api
-    input_tokens = (
-        input_tokens if input_tokens != 0 else usage_data.get("input_tokens", 0)
-    )
-    output_tokens = (
-        output_tokens if output_tokens != 0 else usage_data.get("output_tokens", 0)
-    )
-
-    # 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_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)
+    calc_output_msats = round(output_tokens / 1000 * MSATS_PER_1K_OUTPUT_TOKENS, 3)
+    token_based_cost = math.ceil(calc_input_msats + calc_output_msats)
     total_usd = (token_based_cost / 1000.0) * sats_usd_price()
 
     logger.info(
@@ -220,8 +265,8 @@ async def calculate_cost(  # todo: can be sync
         extra={
             "input_tokens": input_tokens,
             "output_tokens": output_tokens,
-            "input_cost_msats": input_msats,
-            "output_cost_msats": output_msats,
+            "input_cost_msats": calc_input_msats,
+            "output_cost_msats": calc_output_msats,
             "total_cost_msats": token_based_cost,
             "total_usd": total_usd,
             "model": response_data.get("model", "unknown"),
@@ -230,8 +275,10 @@ async def calculate_cost(  # todo: can be sync
 
     return CostData(
         base_msats=0,
-        input_msats=int(input_msats),
-        output_msats=int(output_msats),
+        input_msats=int(calc_input_msats),
+        output_msats=int(calc_output_msats),
         total_msats=token_based_cost,
         total_usd=total_usd,
+        input_tokens=input_tokens,
+        output_tokens=output_tokens,
     )

routstr/payment/fee_schedule.py

@@ -0,0 +1,113 @@
+"""Dynamic provider fee schedule logic.
+
+Supports time-based fee ranges (HH:MM UTC) with overlap validation and active fee resolution.
+"""
+
+from __future__ import annotations
+
+import re
+from datetime import datetime, timezone
+
+from pydantic.v1 import BaseModel, validator
+
+_HH_MM_RE = re.compile(r"^([01]\d|2[0-3]):([0-5]\d)$")
+
+
+class FeeTimeRange(BaseModel):
+    start_time: str  # HH:MM UTC
+    end_time: str  # HH:MM UTC
+    provider_fee: float
+
+    @validator("start_time", "end_time")
+    @classmethod
+    def validate_time_format(cls, v: str) -> str:
+        if not _HH_MM_RE.match(v):
+            raise ValueError(f"Time must be in HH:MM format (00:00–23:59), got: {v!r}")
+        return v
+
+    @validator("provider_fee")
+    @classmethod
+    def validate_fee(cls, v: float) -> float:
+        if v <= 0:
+            raise ValueError(f"provider_fee must be > 0 (got {v})")
+        return v
+
+
+def _to_minutes(t: str) -> int:
+    h, m = map(int, t.split(":"))
+    return h * 60 + m
+
+
+def _range_intervals(r: FeeTimeRange) -> list[tuple[int, int]]:
+    """Return list of [start, end) minute intervals for this range.
+
+    Handles midnight-crossing (e.g. 22:00–06:00 → [(1320,1440),(0,360)]).
+    start == end is treated as a full-day range.
+    """
+    start = _to_minutes(r.start_time)
+    end = _to_minutes(r.end_time)
+    if start < end:
+        return [(start, end)]
+    if start > end:
+        return [(start, 1440), (0, end)]
+    # start == end → full day
+    return [(0, 1440)]
+
+
+def _intervals_overlap(a: tuple[int, int], b: tuple[int, int]) -> bool:
+    return a[0] < b[1] and b[0] < a[1]
+
+
+def ranges_overlap(a: FeeTimeRange, b: FeeTimeRange) -> bool:
+    """Return True if two fee time ranges overlap at any point in the day."""
+    for ia in _range_intervals(a):
+        for ib in _range_intervals(b):
+            if _intervals_overlap(ia, ib):
+                return True
+    return False
+
+
+def validate_no_overlaps(ranges: list[FeeTimeRange]) -> None:
+    """Raise ValueError if any two ranges in the list overlap."""
+    for i in range(len(ranges)):
+        for j in range(i + 1, len(ranges)):
+            if ranges_overlap(ranges[i], ranges[j]):
+                raise ValueError(
+                    f"Fee ranges overlap: [{ranges[i].start_time}–{ranges[i].end_time}]"
+                    f" and [{ranges[j].start_time}–{ranges[j].end_time}]"
+                )
+
+
+def get_active_fee(
+    ranges: list[FeeTimeRange] | None,
+    default_fee: float,
+    *,
+    _now: datetime | None = None,
+) -> float:
+    """Return the provider fee for the current UTC time.
+
+    Falls back to *default_fee* when no range matches or *ranges* is empty/None.
+    The *_now* parameter is for testing only.
+    """
+    if not ranges or not isinstance(ranges, list):
+        return default_fee
+
+    now = _now if _now is not None else datetime.now(timezone.utc)
+    # Normalize to UTC
+    if now.tzinfo is not None:
+        now = now.astimezone(timezone.utc)
+    current = now.hour * 60 + now.minute
+
+    for r in ranges:
+        start = _to_minutes(r.start_time)
+        end = _to_minutes(r.end_time)
+        if start < end:
+            if start <= current < end:
+                return r.provider_fee
+        elif start > end:  # midnight-crossing
+            if current >= start or current < end:
+                return r.provider_fee
+        else:  # full day (start == end)
+            return r.provider_fee
+
+    return default_fee

routstr/payment/helpers.py

@@ -29,15 +29,13 @@ def check_token_balance(headers: dict, body: dict, max_cost_for_model: int) -> N
             },
         )
     elif auth := headers.get("authorization", None):
-        cashu_token = auth.split(" ")[1] if len(auth.split(" ")) > 1 else ""
         logger.debug(
-            "Using Authorization header token",
+            "Skipping preflight token balance check for Authorization header",
             extra={
-                "token_preview": cashu_token[:20] + "..."
-                if len(cashu_token) > 20
-                else cashu_token
+                "auth_preview": auth[:20] + "..." if len(auth) > 20 else auth,
             },
         )
+        return
     else:
         logger.error("No authentication token provided")
         raise HTTPException(status_code=401, detail="Unauthorized")
@@ -75,7 +73,7 @@ def check_token_balance(headers: dict, body: dict, max_cost_for_model: int) -> N
 
     if max_cost_for_model > amount_msat:
         raise HTTPException(
-            status_code=413,
+            status_code=402,
             detail={
                 "reason": "Insufficient balance",
                 "amount_required_msat": max_cost_for_model,
@@ -169,9 +167,32 @@ async def calculate_discounted_max_cost(
 
     tol = settings.tolerance_percentage
     tol_factor = max(0.0, 1 - float(tol) / 100.0)
+
     max_prompt_allowed_sats = model_pricing.max_prompt_cost * tol_factor
     max_completion_allowed_sats = model_pricing.max_completion_cost * tol_factor
 
+    if model_obj:
+        prompt_token_limit: int | None = None
+        if model_obj.top_provider and (
+            model_obj.top_provider.context_length
+            or model_obj.top_provider.max_completion_tokens
+        ):
+            cl = model_obj.top_provider.context_length
+            mct = model_obj.top_provider.max_completion_tokens
+            if cl and mct:
+                prompt_token_limit = max(0, cl - mct)
+            elif cl:
+                prompt_token_limit = cl
+            elif mct:
+                prompt_token_limit = 0
+        elif model_obj.context_length:
+            prompt_token_limit = model_obj.context_length
+
+        if prompt_token_limit is not None:
+            max_prompt_allowed_sats = (
+                prompt_token_limit * model_pricing.prompt * tol_factor
+            )
+
     adjusted = max_cost_for_model
 
     if messages := body.get("messages"):

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.
 
@@ -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

@@ -4,10 +4,11 @@ import random
 
 import httpx
 from fastapi import APIRouter, Depends
+from pydantic import BaseModel as V2BaseModel
 from pydantic.v1 import BaseModel
 from sqlmodel.ext.asyncio.session import AsyncSession
 
-from ..core.db import ModelRow, get_session
+from ..core.db import ModelRow, UpstreamProviderRow, get_session
 from ..core.logging import get_logger
 from ..core.settings import settings
 from .price import sats_usd_price
@@ -60,6 +61,7 @@ class Model(BaseModel):
     upstream_provider_id: int | str | None = None
     canonical_slug: str | None = None
     alias_ids: list[str] | None = None
+    forwarded_model_id: str | None = None
 
     def __hash__(self) -> int:
         return hash(self.id)
@@ -143,14 +145,6 @@ 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:
@@ -185,6 +179,7 @@ def _row_to_model(
         upstream_provider_id=row.upstream_provider_id,
         canonical_slug=getattr(row, "canonical_slug", None),
         alias_ids=json.loads(row.alias_ids) if row.alias_ids else None,
+        forwarded_model_id=getattr(row, "forwarded_model_id", None) or row.id,
     )
 
     if apply_provider_fee:
@@ -203,29 +198,6 @@ 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,
@@ -262,21 +234,6 @@ async def list_models(
     ]
 
 
-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)
-    if not provider or not provider.enabled:
-        return None
-    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.
 
@@ -375,6 +332,7 @@ def _update_model_sats_pricing(model: Model, sats_to_usd: float) -> Model:
             upstream_provider_id=model.upstream_provider_id,
             canonical_slug=model.canonical_slug,
             alias_ids=model.alias_ids,
+            forwarded_model_id=model.forwarded_model_id,
         )
     except Exception as e:
         logger.error(
@@ -448,11 +406,89 @@ async def update_sats_pricing() -> None:
             logger.error(f"Error updating sats pricing: {e}")
 
 
+class ModelTestRequest(V2BaseModel):
+    model_id: str
+    endpoint_type: str
+    request_data: dict
+
+
+@models_router.post("/api/models/test")
+async def test_model(
+    payload: ModelTestRequest,
+    session: AsyncSession = Depends(get_session),
+) -> dict:
+    """Test a model by sending a request through its configured upstream provider."""
+    from sqlmodel import select
+
+    result = await session.execute(
+        select(ModelRow).where(ModelRow.id == payload.model_id)
+    )
+    model_row = result.scalars().first()
+
+    if not model_row:
+        return {
+            "success": False,
+            "error": f"Model '{payload.model_id}' not found in database",
+            "status_code": 404,
+        }
+
+    provider = await session.get(UpstreamProviderRow, model_row.upstream_provider_id)
+    if not provider:
+        return {
+            "success": False,
+            "error": "Upstream provider not found",
+            "status_code": 404,
+        }
+
+    base_url = provider.base_url.rstrip("/")
+    if payload.endpoint_type == "chat-completions":
+        url = f"{base_url}/chat/completions"
+    else:
+        url = f"{base_url}/{payload.endpoint_type}"
+
+    actual_model_id = model_row.forwarded_model_id or model_row.id
+    request_data = dict(payload.request_data)
+    request_data["model"] = actual_model_id
+
+    headers = {
+        "Content-Type": "application/json",
+        "Authorization": f"Bearer {provider.api_key}",
+    }
+
+    try:
+        async with httpx.AsyncClient(timeout=30.0) as client:
+            response = await client.post(url, json=request_data, headers=headers)
+            try:
+                response_data = response.json()
+            except Exception:
+                response_data = {"raw": response.text}
+
+            return {
+                "success": response.status_code < 400,
+                "data": response_data,
+                "status_code": response.status_code,
+            }
+    except Exception as e:
+        return {
+            "success": False,
+            "error": str(e),
+            "status_code": 500,
+        }
+
+
 @models_router.get("/v1/models")
-@models_router.get("/models", include_in_schema=False)
+@models_router.get("/v1/models/", include_in_schema=False)
+@models_router.get("/models")
+@models_router.get("/models/", include_in_schema=False)
 async def models(session: AsyncSession = Depends(get_session)) -> dict:
     """Get all available models from all providers with database overrides applied."""
     from ..proxy import get_unique_models
 
     items = get_unique_models()
-    return {"data": items}
+    data = []
+    for model in items:
+        m = model.dict()
+        if model.forwarded_model_id:
+            m["id"] = model.forwarded_model_id
+        data.append(m)
+    return {"data": data}

routstr/proxy.py

@@ -17,6 +17,7 @@ from .core.db import (
     get_session,
 )
 from .core.exceptions import UpstreamError
+from .core.settings import settings
 from .payment.helpers import (
     calculate_discounted_max_cost,
     check_token_balance,
@@ -43,6 +44,7 @@ async def initialize_upstreams() -> None:
     global _upstreams
     _upstreams = await init_upstreams()
     logger.info(f"Initialized {len(_upstreams)} upstream providers")
+    await sync_provider_fees()
     await refresh_model_maps()
 
 
@@ -54,6 +56,7 @@ async def reinitialize_upstreams() -> None:
         "Re-initialized upstream providers from admin action",
         extra={"provider_count": len(_upstreams)},
     )
+    await sync_provider_fees()
     await refresh_model_maps()
 
 
@@ -68,7 +71,25 @@ def get_upstreams() -> list[BaseUpstreamProvider]:
 
 def get_model_instance(model_id: str) -> Model | None:
     """Get Model instance by ID from global cache."""
-    return _model_instances.get(model_id.lower())
+    if not model_id:
+        return None
+
+    model_id_lower = model_id.lower()
+    # Try exact match first
+    if model := _model_instances.get(model_id_lower):
+        return model
+
+    # Try stripping common version suffixes (e.g., -20251222)
+    # This handles cases where upstream returns a specific version
+    # but we only track the base model name.
+    import re
+
+    base_model_id = re.sub(r"-\d{8}$", "", model_id_lower)
+    if base_model_id != model_id_lower:
+        if model := _model_instances.get(base_model_id):
+            return model
+
+    return None
 
 
 def get_provider_for_model(model_id: str) -> list[BaseUpstreamProvider] | None:
@@ -99,6 +120,12 @@ async def refresh_model_maps() -> None:
     disabled_model_ids: set[str] = set()
 
     for provider in provider_rows:
+        # Match with instance in _upstreams to update its state from DB
+        for upstream in _upstreams:
+            if getattr(upstream, "db_id", None) == provider.id:
+                # This updates fee and merges DB models WITHOUT hitting network
+                await upstream.refresh_models_cache(skip_network=True)
+
         if not provider.enabled:
             continue
         for model in provider.models:
@@ -114,6 +141,39 @@ async def refresh_model_maps() -> None:
     )
 
 
+async def sync_provider_fees() -> None:
+    """Update active provider_fee in database based on schedules and defaults."""
+    from .payment.fee_schedule import FeeTimeRange, get_active_fee
+
+    async with create_session() as session:
+        result = await session.exec(select(UpstreamProviderRow))
+        provider_rows = result.all()
+
+        updated = False
+        for p in provider_rows:
+            schedules = None
+            if p.provider_fee_schedules:
+                try:
+                    schedules = [
+                        FeeTimeRange(**s) for s in json.loads(p.provider_fee_schedules)
+                    ]
+                except Exception:
+                    pass
+
+            active_fee = get_active_fee(schedules, p.provider_fee_default)
+            if p.provider_fee != active_fee:
+                logger.info(
+                    f"Updating active fee for provider {p.id}: {p.provider_fee} -> {active_fee}",
+                    extra={"provider_id": p.id, "active_fee": active_fee},
+                )
+                p.provider_fee = active_fee
+                session.add(p)
+                updated = True
+
+        if updated:
+            await session.commit()
+
+
 async def refresh_model_maps_periodically() -> None:
     """Background task to refresh model maps every minute."""
     import asyncio
@@ -121,6 +181,7 @@ async def refresh_model_maps_periodically() -> None:
     while True:
         try:
             await asyncio.sleep(60)
+            await sync_provider_fees()
             await refresh_model_maps()
         except asyncio.CancelledError:
             break
@@ -137,11 +198,6 @@ async def proxy(
 ) -> Response | StreamingResponse:
     headers = dict(request.headers)
 
-    if "x-cashu" not in headers and "authorization" not in headers.keys():
-        return create_error_response(
-            "unauthorized", "Unauthorized", 401, request=request
-        )
-
     is_responses_api = path.startswith("v1/responses") or path.startswith("responses")
     request_body = await request.body()
     request_body_dict = parse_request_body_json(request_body, path)
@@ -152,6 +208,7 @@ async def proxy(
         model_id = request_body_dict.get("model", "unknown")
 
     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
@@ -176,6 +233,9 @@ async def proxy(
     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):
@@ -206,7 +266,9 @@ async def proxy(
         )
 
     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, model_id
+        )
 
     else:
         if request.method not in ["GET"]:
@@ -271,28 +333,46 @@ async def proxy(
         headers = upstream.prepare_headers(dict(request.headers))
 
         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,
+            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 UpstreamError:
+                # Let the outer UpstreamError handler manage retry/revert
+                raise
+            except Exception as e:
+                # Unexpected error (not an upstream failure) — revert and propagate
+                logger.error(
+                    "Unexpected error in upstream request, reverting payment",
+                    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:
                 # Check if we should retry (502 Upstream Error or 429 Rate Limit)
@@ -367,10 +447,16 @@ async def proxy(
 
 
 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,
+    model_id: str = "unknown",
 ) -> ApiKey:
     """Handle bearer token authentication proxy requests."""
-    bearer_key = auth.replace("Bearer ", "") if auth.startswith("Bearer ") else ""
+    parts = auth.split()
+    bearer_key = parts[1] if len(parts) > 1 and parts[0].lower() == "bearer" else ""
     refund_address = headers.get("Refund-LNURL", None)
     key_expiry_time = headers.get("Key-Expiry-Time", None)
 
@@ -383,6 +469,7 @@ async def get_bearer_token_key(
             "bearer_key_preview": bearer_key[:20] + "..."
             if len(bearer_key) > 20
             else bearer_key,
+            "min_cost": min_cost,
         },
     )
 
@@ -421,6 +508,7 @@ async def get_bearer_token_key(
             session,
             refund_address,
             key_expiry_time,  # type: ignore
+            min_cost=min_cost,
         )
         logger.info(
             "Bearer token validated successfully",
@@ -432,15 +520,16 @@ async def get_bearer_token_key(
         )
         return key
     except Exception as e:
+        key_preview = bearer_key[:20] + "..." if len(bearer_key) > 20 else bearer_key
         logger.error(
-            "Bearer token validation failed",
+            f"Bearer token validation failed: {type(e).__name__}: {e} path={path} model={model_id!r} min_cost={min_cost} key={key_preview!r}",
             extra={
                 "error": str(e),
                 "error_type": type(e).__name__,
                 "path": path,
-                "bearer_key_preview": bearer_key[:20] + "..."
-                if len(bearer_key) > 20
-                else bearer_key,
+                "model_id": model_id,
+                "min_cost_msat": min_cost,
+                "bearer_key_preview": key_preview,
             },
         )
         raise

routstr/upstream/__init__.py

@@ -10,6 +10,7 @@ from .openai import OpenAIUpstreamProvider
 from .openrouter import OpenRouterUpstreamProvider
 from .perplexity import PerplexityUpstreamProvider
 from .ppqai import PPQAIUpstreamProvider
+from .routstr import RoutstrUpstreamProvider
 from .xai import XAIUpstreamProvider
 
 upstream_provider_classes: list[type[BaseUpstreamProvider]] = [
@@ -24,6 +25,7 @@ upstream_provider_classes: list[type[BaseUpstreamProvider]] = [
     OpenRouterUpstreamProvider,
     PerplexityUpstreamProvider,
     PPQAIUpstreamProvider,
+    RoutstrUpstreamProvider,
     XAIUpstreamProvider,
 ]
 """List of all upstream classes"""

routstr/upstream/anthropic.py

@@ -13,6 +13,8 @@ class AnthropicUpstreamProvider(BaseUpstreamProvider):
     provider_type = "anthropic"
     default_base_url = "https://api.anthropic.com/v1"
     platform_url = "https://console.anthropic.com/settings/keys"
+    supports_anthropic_messages = True
+    litellm_provider_prefix = "anthropic/"
 
     def __init__(self, api_key: str, provider_fee: float = 1.01):
         super().__init__(

routstr/upstream/auto_topup.py

@@ -0,0 +1,157 @@
+import asyncio
+import json
+
+from sqlmodel import select
+
+from ..core import get_logger
+from ..core.db import UpstreamProviderRow, create_session
+from ..wallet import send_token
+from .routstr import RoutstrUpstreamProvider
+
+logger = get_logger(__name__)
+
+# Check every 60 seconds
+AUTO_TOPUP_INTERVAL_SECONDS = 60
+
+
+async def periodic_auto_topup() -> None:
+    """Background task that monitors Routstr provider balances and auto-tops up when below threshold.
+
+    For each Routstr provider with auto_topup enabled in provider_settings:
+    1. Checks the upstream balance via get_balance()
+    2. If balance < topup_threshold, creates a cashu token from the configured mint
+    3. Sends the token to the upstream provider via topup()
+    """
+    # Wait for initial startup to complete
+    await asyncio.sleep(30)
+    logger.info("Auto top-up worker started")
+
+    while True:
+        try:
+            await _run_auto_topup_cycle()
+        except Exception as e:
+            logger.error(
+                "Auto top-up cycle failed",
+                extra={"error": str(e), "error_type": type(e).__name__},
+            )
+
+        await asyncio.sleep(AUTO_TOPUP_INTERVAL_SECONDS)
+
+
+async def _run_auto_topup_cycle() -> None:
+    """Single cycle: check all eligible providers and top up if needed."""
+    async with create_session() as session:
+        query = select(UpstreamProviderRow).where(
+            UpstreamProviderRow.provider_type == "routstr",
+            UpstreamProviderRow.enabled == True,  # noqa: E712
+        )
+        result = await session.exec(query)
+        providers = result.all()
+
+    for row in providers:
+        try:
+            await _check_and_topup(row)
+        except Exception as e:
+            logger.error(
+                "Auto top-up failed for provider",
+                extra={
+                    "provider_id": row.id,
+                    "base_url": row.base_url,
+                    "error": str(e),
+                    "error_type": type(e).__name__,
+                },
+            )
+
+
+async def _check_and_topup(row: UpstreamProviderRow) -> None:
+    """Check a single provider's balance and top up if below threshold."""
+    # Parse provider settings
+    settings: dict = {}
+    if row.provider_settings:
+        try:
+            settings = json.loads(row.provider_settings)
+        except (json.JSONDecodeError, TypeError):
+            return
+
+    if not settings.get("auto_topup"):
+        return
+
+    threshold = settings.get("topup_threshold")
+    amount = settings.get("topup_amount_limit")
+    mint_url = settings.get("topup_mint_url")
+
+    if not threshold or not amount or not mint_url:
+        logger.warning(
+            "Auto top-up enabled but missing configuration",
+            extra={
+                "provider_id": row.id,
+                "has_threshold": bool(threshold),
+                "has_amount": bool(amount),
+                "has_mint": bool(mint_url),
+            },
+        )
+        return
+
+    if not row.api_key:
+        return
+
+    # Instantiate provider and check balance
+    provider = RoutstrUpstreamProvider.from_db_row(row)
+    balance = await provider.get_balance()
+
+    if balance is None:
+        logger.warning(
+            "Could not fetch balance for auto top-up",
+            extra={"provider_id": row.id, "base_url": row.base_url},
+        )
+        return
+
+    if balance >= threshold * 1000:
+        return
+
+    # Balance is below threshold - create token and top up
+    logger.info(
+        "Auto top-up triggered",
+        extra={
+            "provider_id": row.id,
+            "balance": balance,
+            "threshold": threshold,
+            "topup_amount": amount,
+            "mint_url": mint_url,
+        },
+    )
+
+    print(amount, mint_url)
+    try:
+        token = await send_token(amount, "sat", mint_url)
+    except Exception as e:
+        logger.error(
+            "Failed to create cashu token for auto top-up",
+            extra={
+                "provider_id": row.id,
+                "amount": amount,
+                "mint_url": mint_url,
+                "error": str(e),
+            },
+        )
+        return
+
+    result = await provider.topup(token)
+
+    if "error" in result:
+        logger.error(
+            "Auto top-up upstream call failed",
+            extra={
+                "provider_id": row.id,
+                "error": result["error"],
+            },
+        )
+    else:
+        logger.info(
+            "Auto top-up completed successfully",
+            extra={
+                "provider_id": row.id,
+                "amount": amount,
+                "new_balance_approx": balance + amount,
+            },
+        )

routstr/upstream/azure.py

@@ -4,6 +4,7 @@ from .base import BaseUpstreamProvider
 
 if TYPE_CHECKING:
     from ..core.db import UpstreamProviderRow
+    from ..payment.models import Model
 
 
 class AzureUpstreamProvider(BaseUpstreamProvider):
@@ -12,6 +13,7 @@ class AzureUpstreamProvider(BaseUpstreamProvider):
     provider_type = "azure"
     default_base_url = None
     platform_url = "https://portal.azure.com/"
+    litellm_provider_prefix = "azure/"
 
     def __init__(
         self,
@@ -58,19 +60,52 @@ class AzureUpstreamProvider(BaseUpstreamProvider):
             "platform_url": cls.platform_url,
         }
 
+    def prepare_headers(self, request_headers: dict) -> dict:
+        """Prepare headers for Azure OpenAI, adding api-key."""
+        headers = super().prepare_headers(request_headers)
+        if self.api_key:
+            headers["api-key"] = self.api_key
+            headers.pop("Authorization", None)
+            headers.pop("authorization", None)
+        return headers
+
     def prepare_params(
         self, path: str, query_params: Mapping[str, str] | None
     ) -> Mapping[str, str]:
-        """Prepare query parameters for Azure OpenAI, adding API version.
-
-        Args:
-            path: Request path
-            query_params: Original query parameters from the client
-
-        Returns:
-            Query parameters dict with Azure API version added for chat completions
-        """
+        """Prepare query parameters for Azure OpenAI, adding API version."""
         params = dict(query_params or {})
-        if path.endswith("chat/completions"):
-            params["api-version"] = self.api_version
+        version = (self.api_version or "").replace("\ufeff", "").strip()
+        if not version or version.lower() == "v1":
+            version = "2024-02-15-preview"
+        params["api-version"] = version
         return params
+
+    def normalize_request_path(
+        self, path: str, model_obj: "Model | None" = None
+    ) -> str:
+        """Build Azure deployment-specific request path."""
+        clean_path = super().normalize_request_path(path, model_obj).lstrip("/")
+        if model_obj is None:
+            return clean_path
+
+        deployment_id = getattr(
+            model_obj, "canonical_slug", None
+        ) or self.transform_model_name(model_obj.id)
+        deployment_id = deployment_id.split("/")[-1]
+        return f"openai/deployments/{deployment_id}/{clean_path}"
+
+    def get_request_base_url(
+        self, path: str, model_obj: "Model | None" = None
+    ) -> str:
+        """Use endpoint root, stripping accidental /openai/v1 suffix if present."""
+        base_url = self.base_url.rstrip("/")
+        marker = "/openai/v1"
+        if marker in base_url:
+            base_url = base_url.split(marker, 1)[0].rstrip("/")
+        return base_url
+
+    def transform_model_name(self, model_id: str) -> str:
+        """Extract deployment name from model ID."""
+        if "/" in model_id:
+            return model_id.split("/")[-1]
+        return model_id

routstr/upstream/clients/gemini.py

@@ -26,14 +26,14 @@ class GeminiClient(BaseAPIClient):
         max_tokens: int | None = None,
         **kwargs: Any,
     ) -> dict[str, Any]:
-        from openai import NOT_GIVEN
+        from openai import omit
 
         response = await self.client.chat.completions.create(
             model=model,
             messages=messages,  # type: ignore
-            temperature=temperature if temperature is not None else NOT_GIVEN,
-            max_tokens=max_tokens if max_tokens is not None else NOT_GIVEN,
-            top_p=kwargs.get("top_p", NOT_GIVEN),
+            temperature=temperature if temperature is not None else omit,
+            max_tokens=max_tokens if max_tokens is not None else omit,
+            top_p=kwargs.get("top_p", omit),
         )
         return response.model_dump()
 
@@ -45,7 +45,7 @@ class GeminiClient(BaseAPIClient):
         max_tokens: int | None = None,
         **kwargs: Any,
     ) -> AsyncGenerator[dict[str, Any], None]:
-        from openai import NOT_GIVEN
+        from openai import omit
 
         usage_callback = kwargs.get("usage_callback")
         completion_callback = kwargs.get("completion_callback")
@@ -55,9 +55,9 @@ class GeminiClient(BaseAPIClient):
             messages=messages,  # type: ignore
             stream=True,
             stream_options={"include_usage": True},
-            temperature=temperature if temperature is not None else NOT_GIVEN,
-            max_tokens=max_tokens if max_tokens is not None else NOT_GIVEN,
-            top_p=kwargs.get("top_p", NOT_GIVEN),
+            temperature=temperature if temperature is not None else omit,
+            max_tokens=max_tokens if max_tokens is not None else omit,
+            top_p=kwargs.get("top_p", omit),
         )
 
         final_usage = None

routstr/upstream/fireworks.py

@@ -12,6 +12,7 @@ class FireworksUpstreamProvider(BaseUpstreamProvider):
     provider_type = "fireworks"
     default_base_url = "https://api.fireworks.ai/inference/v1"
     platform_url = "https://app.fireworks.ai/settings/users/api-keys"
+    litellm_provider_prefix = "fireworks_ai/"
 
     def __init__(self, api_key: str, provider_fee: float = 1.01):
         super().__init__(

routstr/upstream/gemini.py

@@ -23,6 +23,7 @@ class GeminiUpstreamProvider(BaseUpstreamProvider):
     provider_type = "gemini"
     default_base_url = "https://generativelanguage.googleapis.com/v1beta"
     platform_url = "https://aistudio.google.com/app/apikey"
+    litellm_provider_prefix = "gemini/"
 
     def __init__(
         self,
@@ -259,7 +260,15 @@ class GeminiUpstreamProvider(BaseUpstreamProvider):
                 cost_data = await adjust_payment_for_tokens(
                     key, openai_format_response, session, max_cost_for_model
                 )
+                await session.refresh(key)
+                remaining_balance_msats = key.balance
                 openai_format_response["cost"] = cost_data
+                openai_format_response["cost"]["sats_cost"] = (
+                    cost_data.get("total_msats", 0) // 1000
+                )
+                openai_format_response["cost"]["remaining_balance_msats"] = (
+                    remaining_balance_msats
+                )
 
                 logger.info(
                     "Gemini non-streaming payment completed",

routstr/upstream/groq.py

@@ -12,6 +12,7 @@ class GroqUpstreamProvider(BaseUpstreamProvider):
     provider_type = "groq"
     default_base_url = "https://api.groq.com/openai/v1"
     platform_url = "https://console.groq.com/keys"
+    litellm_provider_prefix = "groq/"
 
     def __init__(self, api_key: str, provider_fee: float = 1.01):
         super().__init__(

routstr/upstream/helpers.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 import asyncio
 import os
 import re
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Callable
 
 if TYPE_CHECKING:
     from ..core.settings import Settings
@@ -122,12 +122,16 @@ async def get_all_models_with_overrides(
 
 
 async def refresh_upstreams_models_periodically(
-    upstreams: list[BaseUpstreamProvider],
+    upstreams_provider: (
+        Callable[[], list[BaseUpstreamProvider]] | list[BaseUpstreamProvider]
+    ),
 ) -> None:
     """Background task to periodically refresh models cache for all providers.
 
     Args:
-        upstreams: List of upstream provider instances
+        upstreams_provider: Either a callable returning the live upstream list
+            (preferred — picks up providers added/changed via reinitialize_upstreams),
+            or a static list (legacy, will go stale after reinitialize_upstreams).
     """
     import asyncio
     import random
@@ -139,9 +143,14 @@ async def refresh_upstreams_models_periodically(
         logger.info("Provider models refresh disabled (interval <= 0)")
         return
 
+    def _resolve_upstreams() -> list[BaseUpstreamProvider]:
+        if callable(upstreams_provider):
+            return upstreams_provider()
+        return upstreams_provider
+
     while True:
         try:
-            for upstream in upstreams:
+            for upstream in _resolve_upstreams():
                 try:
                     await upstream.refresh_models_cache()
                 except Exception as e:
@@ -205,6 +214,9 @@ async def init_upstreams() -> list[BaseUpstreamProvider]:
 
             provider = _instantiate_provider(provider_row)
             if provider:
+                # Keep provider DB id on runtime instance so model mapping can
+                # bind DB overrides to the correct upstream.
+                setattr(provider, "db_id", provider_row.id)
                 await provider.refresh_models_cache()
                 logger.debug(
                     f"Initialized {provider_row.provider_type} provider",

routstr/upstream/ollama.py

@@ -3,13 +3,11 @@ from __future__ import annotations
 from typing import TYPE_CHECKING
 
 import httpx
-from fastapi import Request
-from fastapi.responses import Response, StreamingResponse
 
 from .base import BaseUpstreamProvider
 
 if TYPE_CHECKING:
-    from ..core.db import ApiKey, AsyncSession, UpstreamProviderRow
+    from ..core.db import UpstreamProviderRow
     from ..payment.models import Model
 
 from ..core.logging import get_logger
@@ -23,6 +21,7 @@ class OllamaUpstreamProvider(BaseUpstreamProvider):
     provider_type = "ollama"
     default_base_url = "http://localhost:11434"
     platform_url = None
+    litellm_provider_prefix = "ollama_chat/"
 
     def __init__(
         self,
@@ -67,38 +66,9 @@ class OllamaUpstreamProvider(BaseUpstreamProvider):
         """Strip 'ollama/' prefix for Ollama API compatibility."""
         return model_id.removeprefix("ollama/")
 
-    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:
-        """Override to use OpenAI-compatible endpoint for proxy requests."""
-        if path.startswith("v1/"):
-            path = path.replace("v1/", "")
-
-        original_base_url = self.base_url
-        self.base_url = f"{self.base_url}/v1"
-
-        try:
-            result = await super().forward_request(
-                request,
-                path,
-                headers,
-                request_body,
-                key,
-                max_cost_for_model,
-                session,
-                model_obj,
-            )
-            return result
-        finally:
-            self.base_url = original_base_url
+    def get_request_base_url(self, path: str, model_obj: Model | None = None) -> str:
+        """Route proxy traffic through Ollama's OpenAI-compatible /v1 endpoint."""
+        return f"{self.base_url.rstrip('/')}/v1"
 
     async def fetch_models(self) -> list[Model]:
         """Fetch models from Ollama API using /api/tags endpoint."""
@@ -195,103 +165,3 @@ class OllamaUpstreamProvider(BaseUpstreamProvider):
                 },
             )
             return []
-
-    async def refresh_models_cache(self) -> None:
-        """Refresh the in-memory models cache from upstream API."""
-        try:
-            from ..payment.models import _update_model_sats_pricing
-            from ..payment.price import sats_usd_price
-
-            models = await self.fetch_models()
-            models_with_fees = [self._apply_provider_fee_to_model(m) for m in models]
-
-            try:
-                sats_to_usd = sats_usd_price()
-                self._models_cache = [
-                    _update_model_sats_pricing(m, sats_to_usd) for m in models_with_fees
-                ]
-            except Exception:
-                self._models_cache = models_with_fees
-
-            self._models_by_id = {m.id: m for m in self._models_cache}
-            logger.info(
-                f"Refreshed models cache for {self.base_url}",
-                extra={"model_count": len(models)},
-            )
-        except Exception as e:
-            logger.error(
-                f"Failed to refresh models cache for {self.base_url}",
-                extra={"error": str(e), "error_type": type(e).__name__},
-            )
-
-    def get_cached_models(self) -> list[Model]:
-        """Get cached models for this provider.
-
-        Returns:
-            List of cached Model objects
-        """
-        return self._models_cache
-
-    def get_cached_model_by_id(self, model_id: str) -> Model | None:
-        """Get a specific cached model by ID.
-
-        Args:
-            model_id: Model identifier
-
-        Returns:
-            Model object or None if not found
-        """
-        return self._models_by_id.get(model_id)
-
-    def _apply_provider_fee_to_model(self, model: Model) -> Model:
-        """Apply provider fee to model's USD pricing and calculate max costs.
-
-        Args:
-            model: Model object to update
-
-        Returns:
-            Model with provider fee applied to pricing and max costs calculated
-        """
-        from ..payment.models import Model, Pricing, _calculate_usd_max_costs
-
-        adjusted_pricing = Pricing.parse_obj(
-            {k: v * self.provider_fee for k, v in model.pricing.dict().items()}
-        )
-
-        temp_model = Model(
-            id=model.id,
-            name=model.name,
-            created=model.created,
-            description=model.description,
-            context_length=model.context_length,
-            architecture=model.architecture,
-            pricing=adjusted_pricing,
-            sats_pricing=None,
-            per_request_limits=model.per_request_limits,
-            top_provider=model.top_provider,
-            enabled=model.enabled,
-            upstream_provider_id=model.upstream_provider_id,
-            canonical_slug=model.canonical_slug,
-        )
-
-        (
-            adjusted_pricing.max_prompt_cost,
-            adjusted_pricing.max_completion_cost,
-            adjusted_pricing.max_cost,
-        ) = _calculate_usd_max_costs(temp_model)
-
-        return Model(
-            id=model.id,
-            name=model.name,
-            created=model.created,
-            description=model.description,
-            context_length=model.context_length,
-            architecture=model.architecture,
-            pricing=adjusted_pricing,
-            sats_pricing=model.sats_pricing,
-            per_request_limits=model.per_request_limits,
-            top_provider=model.top_provider,
-            enabled=model.enabled,
-            upstream_provider_id=model.upstream_provider_id,
-            canonical_slug=model.canonical_slug,
-        )

routstr/upstream/openrouter.py

@@ -15,6 +15,8 @@ class OpenRouterUpstreamProvider(BaseUpstreamProvider):
     provider_type = "openrouter"
     default_base_url = "https://openrouter.ai/api/v1"
     platform_url = "https://openrouter.ai/settings/keys"
+    supports_anthropic_messages = True
+    litellm_provider_prefix = "openrouter/"
 
     def __init__(self, api_key: str, provider_fee: float = 1.06):
         """Initialize OpenRouter provider with API key.

routstr/upstream/perplexity.py

@@ -13,6 +13,7 @@ class PerplexityUpstreamProvider(BaseUpstreamProvider):
     provider_type = "perplexity"
     default_base_url = "https://api.perplexity.ai/"
     platform_url = "https://www.perplexity.ai/account/api/keys"
+    litellm_provider_prefix = "perplexity/"
 
     def __init__(self, api_key: str, provider_fee: float = 1.01):
         super().__init__(

routstr/upstream/ppqai.py

@@ -1,9 +1,9 @@
 from __future__ import annotations
 
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Optional
 
 import httpx
-from pydantic import BaseModel
+from pydantic.v1 import BaseModel, Field
 
 from ..core.logging import get_logger
 from ..payment.models import Architecture, Model, Pricing, async_fetch_openrouter_models
@@ -16,18 +16,20 @@ logger = get_logger(__name__)
 
 
 class PPQAIModelPricing(BaseModel):
-    ui: dict[str, float]
-    api: dict[str, float]
+    ui: Optional[dict[str, float]] = None
+    api: Optional[dict[str, float]] = None
+    input_per_1M_tokens: Optional[float] = Field(None, alias="input_per_1M_tokens")
+    output_per_1M_tokens: Optional[float] = Field(None, alias="output_per_1M_tokens")
 
 
 class PPQAIModel(BaseModel):
     id: str
-    provider: str
+    provider: Optional[str] = None
     name: str
     created_at: int
     context_length: int
     pricing: PPQAIModelPricing
-    popular: bool
+    popular: bool = False
 
 
 class PPQAIUpstreamProvider(BaseUpstreamProvider):
@@ -134,31 +136,54 @@ class PPQAIUpstreamProvider(BaseUpstreamProvider):
                         )
 
                         if or_model:
-                            if input_price := ppqai_model.pricing.api.get(
-                                "input_per_1M"
-                            ):
+                            input_price = None
+                            if ppqai_model.pricing.api:
+                                input_price = ppqai_model.pricing.api.get(
+                                    "input_per_1M"
+                                )
+                            elif ppqai_model.pricing.input_per_1M_tokens:
+                                input_price = ppqai_model.pricing.input_per_1M_tokens
+
+                            if input_price is not None:
                                 or_model.pricing.prompt = input_price / 1_000_000
-                            if output_price := ppqai_model.pricing.api.get(
-                                "output_per_1M"
-                            ):
+
+                            output_price = None
+                            if ppqai_model.pricing.api:
+                                output_price = ppqai_model.pricing.api.get(
+                                    "output_per_1M"
+                                )
+                            elif ppqai_model.pricing.output_per_1M_tokens:
+                                output_price = ppqai_model.pricing.output_per_1M_tokens
+
+                            if output_price is not None:
                                 or_model.pricing.completion = output_price / 1_000_000
+
                             if cl := ppqai_model.context_length:
                                 or_model.context_length = cl
                             models.append(or_model)
                         else:
-                            input_price = ppqai_model.pricing.api.get(
-                                "input_per_1M", 0.0
-                            )
-                            output_price = ppqai_model.pricing.api.get(
-                                "output_per_1M", 0.0
-                            )
+                            input_price = 0.0
+                            if ppqai_model.pricing.api:
+                                input_price = ppqai_model.pricing.api.get(
+                                    "input_per_1M", 0.0
+                                )
+                            elif ppqai_model.pricing.input_per_1M_tokens:
+                                input_price = ppqai_model.pricing.input_per_1M_tokens
+
+                            output_price = 0.0
+                            if ppqai_model.pricing.api:
+                                output_price = ppqai_model.pricing.api.get(
+                                    "output_per_1M", 0.0
+                                )
+                            elif ppqai_model.pricing.output_per_1M_tokens:
+                                output_price = ppqai_model.pricing.output_per_1M_tokens
 
                             models.append(
                                 Model(
                                     id=ppqai_model.id,
                                     name=ppqai_model.name,
                                     created=ppqai_model.created_at // 1000,
-                                    description=f"{ppqai_model.provider} model",
+                                    description=f"{ppqai_model.provider or 'PPQ.AI'} model",
                                     context_length=ppqai_model.context_length,
                                     architecture=Architecture(
                                         modality="text->text",
@@ -283,7 +308,6 @@ class PPQAIUpstreamProvider(BaseUpstreamProvider):
         )
 
         async with httpx.AsyncClient(timeout=30.0) as client:
-            print(f"Payload: {payload}", "sending to", url)
             response = await client.post(url, headers=headers, json=payload)
             response.raise_for_status()
             invoice_data = response.json()

routstr/upstream/routstr.py

@@ -0,0 +1,169 @@
+from typing import TYPE_CHECKING, Any
+
+import httpx
+
+from ..core import get_logger
+from ..payment.models import Model
+from .base import BaseUpstreamProvider
+
+if TYPE_CHECKING:
+    from ..core.db import UpstreamProviderRow
+
+logger = get_logger(__name__)
+
+
+class RoutstrUpstreamProvider(BaseUpstreamProvider):
+    """Upstream provider for communicating with another Routstr instance."""
+
+    provider_type = "routstr"
+    default_base_url = None
+    platform_url = None
+
+    def __init__(
+        self,
+        base_url: str,
+        api_key: str,
+        provider_fee: float = 1.01,
+        provider_settings: dict | None = None,
+    ):
+        """Initialize Routstr provider.
+
+        Args:
+            base_url: Base URL of the upstream Routstr instance
+            api_key: API key for the upstream Routstr instance
+            provider_fee: Provider fee multiplier
+            provider_settings: Provider-specific settings (auto-topup, etc.)
+        """
+        # Ensure base_url doesn't end with /v1 as BaseUpstreamProvider appends it if needed
+        # but Routstr paths are usually absolute from base.
+        super().__init__(
+            base_url=base_url.rstrip("/"),
+            api_key=api_key,
+            provider_fee=provider_fee,
+        )
+        self.settings = provider_settings or {}
+
+    @classmethod
+    def from_db_row(
+        cls, provider_row: "UpstreamProviderRow"
+    ) -> "RoutstrUpstreamProvider":
+        import json
+
+        settings = {}
+        if provider_row.provider_settings:
+            try:
+                settings = json.loads(provider_row.provider_settings)
+            except Exception:
+                pass
+
+        return cls(
+            base_url=provider_row.base_url,
+            api_key=provider_row.api_key,
+            provider_fee=provider_row.provider_fee,
+            provider_settings=settings,
+        )
+
+    @classmethod
+    def get_provider_metadata(cls) -> dict[str, object]:
+        return {
+            "id": cls.provider_type,
+            "name": "Routstr Node",
+            "default_base_url": "",
+            "fixed_base_url": False,
+            "platform_url": cls.platform_url,
+            "can_create_account": False,
+            "can_topup": True,
+            "can_show_balance": True,
+        }
+
+    async def get_balance(self) -> float | None:
+        """Fetch balance from the upstream Routstr node.
+
+        Returns:
+            Balance in satoshis, or None if failed
+        """
+        url = f"{self.base_url}/v1/balance/info"
+        headers = {"Authorization": f"Bearer {self.api_key}"} if self.api_key else {}
+
+        async with httpx.AsyncClient() as client:
+            try:
+                response = await client.get(url, headers=headers, timeout=10.0)
+                response.raise_for_status()
+                data = response.json()
+                # Routstr balance info usually contains 'balance' in msats or sats
+                # Check for msats and convert to sats
+                if "balance_msats" in data:
+                    return float(data["balance_msats"]) / 1000.0
+                return float(data.get("balance", 0))
+            except Exception as e:
+                logger.error(
+                    "Failed to fetch balance from upstream Routstr",
+                    extra={"url": url, "error": str(e)},
+                )
+                return None
+
+    async def topup(self, cashu_token: str) -> dict[str, Any]:
+        """Top up balance on the upstream Routstr node.
+
+        Args:
+            cashu_token: Cashu token to deposit
+
+        Returns:
+            Dict containing top-up result
+        """
+        url = f"{self.base_url}/v1/balance/topup"
+        headers = {"Authorization": f"Bearer {self.api_key}"}
+        payload = {"cashu_token": cashu_token}
+
+        async with httpx.AsyncClient() as client:
+            try:
+                response = await client.post(
+                    url, headers=headers, json=payload, timeout=30.0
+                )
+                response.raise_for_status()
+                return response.json()
+            except Exception as e:
+                logger.error(
+                    "Failed to topup upstream Routstr",
+                    extra={"url": url, "error": str(e)},
+                )
+                return {"error": str(e)}
+
+    async def fetch_models(self) -> list[Model]:
+        """Fetch models from the upstream Routstr node."""
+        url = f"{self.base_url}/v1/models"
+
+        async with httpx.AsyncClient() as client:
+            try:
+                response = await client.get(url, headers={}, timeout=15.0)
+                response.raise_for_status()
+                data = response.json()
+                models = data.get("data", [])
+                return [Model(**m) for m in models]
+            except Exception as e:
+                logger.error(
+                    "Failed to fetch models from upstream Routstr",
+                    extra={"url": url, "error": str(e)},
+                )
+                return []
+
+    async def refund_balance(self) -> dict[str, Any]:
+        """Request a refund from the upstream Routstr node.
+
+        Returns:
+            Dict containing refund result and token
+        """
+        url = f"{self.base_url}/v1/balance/refund"
+        headers = {"Authorization": f"Bearer {self.api_key}"}
+
+        async with httpx.AsyncClient() as client:
+            try:
+                response = await client.post(url, headers=headers, timeout=30.0)
+                response.raise_for_status()
+                return response.json()
+            except Exception as e:
+                logger.error(
+                    "Failed to request refund from upstream Routstr",
+                    extra={"url": url, "error": str(e)},
+                )
+                return {"error": str(e)}

routstr/upstream/xai.py

@@ -13,6 +13,7 @@ class XAIUpstreamProvider(BaseUpstreamProvider):
     provider_type = "x-ai"
     default_base_url = "https://api.x.ai/v1"
     platform_url = "https://console.x.ai/"
+    litellm_provider_prefix = "xai/"
 
     def __init__(self, api_key: str, provider_fee: float = 1.01):
         super().__init__(

routstr/wallet.py

@@ -1,16 +1,32 @@
 import asyncio
-import math
+import time
+import typing
 from typing import TypedDict
 
 from cashu.core.base import Proof, Token
+from cashu.core.mint_info import MintInfo as _CashuMintInfo
 from cashu.wallet.helpers import deserialize_token_from_string
 from cashu.wallet.wallet import Wallet
-from sqlmodel import col, update
+from pydantic_core import PydanticUndefined
+from sqlmodel import col, select, update
 
 from .core import db, get_logger
+from .core.db import store_cashu_transaction
 from .core.settings import settings
 from .payment.lnurl import raw_send_to_lnurl
 
+# cashu still declares Optional[X] without explicit defaults on MintInfo.
+# Under pydantic v2 those are required, but real mints omit many of them.
+# Default Optional fields to None at import time so balance fetches don't 422.
+for _name, _field in _CashuMintInfo.model_fields.items():
+    _annot = _field.annotation
+    _is_optional = typing.get_origin(_annot) is typing.Union and type(
+        None
+    ) in typing.get_args(_annot)
+    if _is_optional and _field.default is PydanticUndefined:
+        _field.default = None
+_CashuMintInfo.model_rebuild(force=True)
+
 logger = get_logger(__name__)
 
 
@@ -34,6 +50,7 @@ async def recieve_token(
 
     wallet.verify_proofs_dleq(token_obj.proofs)
     await wallet.split(proofs=token_obj.proofs, amount=0, include_fees=True)
+
     return token_obj.amount, token_obj.unit, token_obj.mint
 
 
@@ -58,15 +75,86 @@ async def send_token(amount: int, unit: str, mint_url: str | None = None) -> str
     return token
 
 
+async def _calculate_swap_amount(
+    amount_msat: int,
+    token_unit: str,
+    token_mint_url: str,
+    token_wallet: Wallet,
+    primary_wallet: Wallet,
+) -> int:
+    """
+    Calculate the amount to mint on the primary mint after accounting for
+    potential swap fees (melt fees) on the foreign mint.
+    """
+    if settings.primary_mint_unit == "sat":
+        receive_amount = amount_msat // 1000
+    else:
+        receive_amount = amount_msat
+
+    if token_mint_url == settings.primary_mint:
+        logger.info(
+            "swap_to_primary_mint: skipping fee estimation (same mint)",
+            extra={"minted_amount": receive_amount},
+        )
+        return int(receive_amount)
+
+    logger.info(
+        "swap_to_primary_mint: estimating fees",
+        extra={
+            "dummy_amount": receive_amount,
+            "unit": settings.primary_mint_unit,
+        },
+    )
+
+    try:
+        dummy_mint_quote = await primary_wallet.request_mint(receive_amount)
+        dummy_melt_quote = await token_wallet.melt_quote(dummy_mint_quote.request)
+
+        fee_reserve = dummy_melt_quote.fee_reserve
+        if token_unit == "sat":
+            fee_msat = fee_reserve * 1000
+        else:
+            fee_msat = fee_reserve
+
+        amount_msat_after_fee = amount_msat - fee_msat
+
+        if settings.primary_mint_unit == "sat":
+            minted_amount = int(amount_msat_after_fee // 1000)
+        else:
+            minted_amount = int(amount_msat_after_fee)
+
+        if minted_amount <= 0:
+            raise ValueError(f"Fees ({fee_reserve} {token_unit}) exceed token amount")
+
+        logger.info(
+            "swap_to_primary_mint: fee estimation result",
+            extra={
+                "token_amount_sat": amount_msat // 1000,
+                "estimated_fee_sat": fee_msat // 1000,
+                "minted_amount": minted_amount,
+                "minted_unit": settings.primary_mint_unit,
+            },
+        )
+        return minted_amount
+
+    except Exception as e:
+        logger.error(
+            "swap_to_primary_mint: fee estimation failed",
+            extra={"error": str(e)},
+        )
+        raise ValueError(f"Failed to estimate fees: {e}") from e
+
+
 async def swap_to_primary_mint(
     token_obj: Token, token_wallet: Wallet
 ) -> tuple[int, str, str]:
     logger.info(
-        "swap_to_primary_mint",
+        "swap_to_primary_mint: starting",
         extra={
-            "mint": token_obj.mint,
-            "amount": token_obj.amount,
+            "foreign_mint": token_obj.mint,
+            "token_amount": token_obj.amount,
             "unit": token_obj.unit,
+            "primary_mint": settings.primary_mint,
         },
     )
     # Ensure amount is an integer
@@ -81,24 +169,118 @@ 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)) + 1
-    amount_msat_after_fee = amount_msat - estimated_fee_sat * 1000
     primary_wallet = await get_wallet(settings.primary_mint, settings.primary_mint_unit)
 
-    if settings.primary_mint_unit == "sat":
-        minted_amount = int(amount_msat_after_fee // 1000)
-    else:
-        minted_amount = int(amount_msat_after_fee)
+    # If the token is already from the primary mint, we don't need to swap
+    # and we definitely don't want to calculate or pay fees.
+    if token_obj.mint == settings.primary_mint:
+        logger.info(
+            "swap_to_primary_mint: token already on primary mint, skipping swap",
+            extra={
+                "mint": token_obj.mint,
+                "amount": token_amount,
+                "unit": token_obj.unit,
+            },
+        )
+        await token_wallet.split(proofs=token_obj.proofs, amount=0, include_fees=True)
+        return token_amount, token_obj.unit, token_obj.mint
+
+    minted_amount = await _calculate_swap_amount(
+        amount_msat,
+        token_obj.unit,
+        token_obj.mint,
+        token_wallet,
+        primary_wallet,
+    )
+
     mint_quote = await primary_wallet.request_mint(minted_amount)
+    logger.info(
+        "swap_to_primary_mint: mint quote received",
+        extra={"mint_quote_id": mint_quote.quote},
+    )
 
     melt_quote = await token_wallet.melt_quote(mint_quote.request)
-    _ = await token_wallet.melt(
-        proofs=token_obj.proofs,
-        invoice=mint_quote.request,
-        fee_reserve_sat=melt_quote.fee_reserve,
-        quote_id=melt_quote.quote,
+    total_needed = melt_quote.amount + melt_quote.fee_reserve
+    logger.info(
+        "swap_to_primary_mint: melt quote received",
+        extra={
+            "melt_quote_id": melt_quote.quote,
+            "melt_amount": melt_quote.amount,
+            "melt_fee_reserve": melt_quote.fee_reserve,
+            "total_needed": total_needed,
+            "token_amount": token_amount,
+        },
+    )
+
+    if total_needed > token_amount:
+        logger.warning(
+            "swap_to_primary_mint: insufficient token amount for melt fees",
+            extra={
+                "token_amount": token_amount,
+                "melt_amount": melt_quote.amount,
+                "melt_fee_reserve": melt_quote.fee_reserve,
+                "total_needed": total_needed,
+                "shortfall": total_needed - token_amount,
+            },
+        )
+        raise ValueError(
+            f"Token amount ({token_amount} {token_obj.unit}) is insufficient to cover "
+            f"melt fees. Needed: {total_needed} {token_obj.unit} "
+            f"(amount: {melt_quote.amount} + fee: {melt_quote.fee_reserve})"
+        )
+
+    try:
+        _ = await token_wallet.melt(
+            proofs=token_obj.proofs,
+            invoice=mint_quote.request,
+            fee_reserve_sat=melt_quote.fee_reserve,
+            quote_id=melt_quote.quote,
+        )
+    except Exception as e:
+        logger.error(
+            "swap_to_primary_mint: melt failed",
+            extra={
+                "error": str(e),
+                "error_type": type(e).__name__,
+                "foreign_mint": token_obj.mint,
+                "token_amount": token_amount,
+                "melt_quote_id": melt_quote.quote,
+                "total_needed": total_needed,
+            },
+        )
+        raise ValueError(
+            f"Failed to melt token from foreign mint {token_obj.mint}: {e}"
+        ) from e
+
+    logger.info(
+        "swap_to_primary_mint: melt succeeded, minting on primary",
+        extra={"minted_amount": minted_amount, "mint_quote_id": mint_quote.quote},
+    )
+
+    try:
+        _ = await primary_wallet.mint(minted_amount, quote_id=mint_quote.quote)
+    except Exception as e:
+        logger.error(
+            "swap_to_primary_mint: mint on primary failed after successful melt",
+            extra={
+                "error": str(e),
+                "error_type": type(e).__name__,
+                "minted_amount": minted_amount,
+                "mint_quote_id": mint_quote.quote,
+            },
+        )
+        raise
+
+    logger.info(
+        "swap_to_primary_mint: completed successfully",
+        extra={
+            "foreign_mint": token_obj.mint,
+            "primary_mint": settings.primary_mint,
+            "original_amount": token_amount,
+            "minted_amount": minted_amount,
+            "unit": settings.primary_mint_unit,
+        },
     )
-    _ = await primary_wallet.mint(minted_amount, quote_id=mint_quote.quote)
 
     return int(minted_amount), settings.primary_mint_unit, settings.primary_mint
 
@@ -113,6 +295,8 @@ async def credit_balance(
 
     try:
         amount, unit, mint_url = await recieve_token(cashu_token)
+        original_amount = amount
+        original_unit = unit
         logger.info(
             "credit_balance: Token redeemed successfully",
             extra={"amount": amount, "unit": unit, "mint_url": mint_url},
@@ -144,6 +328,19 @@ async def credit_balance(
             extra={"new_balance": key.balance},
         )
 
+        try:
+            await store_cashu_transaction(
+                token=cashu_token,
+                amount=original_amount,
+                unit=original_unit,
+                mint_url=mint_url,
+                typ="in",
+                source="apikey",
+                api_key_hashed_key=key.hashed_key,
+            )
+        except Exception:
+            pass
+
         logger.info(
             "Cashu token successfully redeemed and stored",
             extra={"amount": amount, "unit": unit, "mint_url": mint_url},
@@ -319,6 +516,7 @@ async def periodic_payout() -> None:
                             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
                         )
@@ -344,8 +542,6 @@ 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__}",
@@ -353,6 +549,101 @@ async def periodic_payout() -> None:
             )
 
 
+async def periodic_refund_sweep() -> None:
+    while True:
+        await asyncio.sleep(60 * 60)  # every hour
+        try:
+            cutoff = int(time.time()) - settings.refund_sweep_ttl_seconds
+            async with db.create_session() as session:
+                stmt = select(db.CashuTransaction).where(
+                    db.CashuTransaction.type == "out",
+                    db.CashuTransaction.collected == False,  # noqa: E712
+                    db.CashuTransaction.swept == False,  # noqa: E712
+                    db.CashuTransaction.created_at < cutoff,
+                )
+                results = await session.exec(stmt)
+                refunds = results.all()
+
+                for refund in refunds:
+                    try:
+                        await recieve_token(refund.token)
+                        refund.swept = True
+                        session.add(refund)
+                        logger.info(
+                            "Swept uncollected refund",
+                            extra={
+                                "id": refund.id,
+                                "amount": refund.amount,
+                                "unit": refund.unit,
+                            },
+                        )
+                    except Exception as e:
+                        error_msg = str(e).lower()
+                        if "already spent" in error_msg:
+                            refund.collected = True
+                            session.add(refund)
+                            logger.info(
+                                "Refund already spent (client collected), marking swept",
+                                extra={
+                                    "id": refund.id,
+                                },
+                            )
+                        else:
+                            logger.warning(
+                                "Failed to sweep refund",
+                                extra={
+                                    "id": refund.id,
+                                    "error": str(e),
+                                },
+                            )
+                await session.commit()
+        except Exception as e:
+            logger.error(
+                "Error in periodic refund sweep",
+                extra={"error": str(e), "error_type": type(e).__name__},
+            )
+
+
+async def periodic_routstr_fee_payout() -> None:
+    from .auth import (
+        ROUTSTR_FEE_DEFAULT_PAYOUT,
+        ROUTSTR_FEE_PAYOUT_INTERVAL_SECONDS,
+        ROUTSTR_LN_ADDRESS,
+    )
+
+    if not ROUTSTR_LN_ADDRESS:
+        logger.info("ROUTSTR_LN_ADDRESS not set, skipping fee payout")
+        return
+    while True:
+        await asyncio.sleep(ROUTSTR_FEE_PAYOUT_INTERVAL_SECONDS)
+        try:
+            async with db.create_session() as session:
+                fee = await db.get_routstr_fee(session)
+                accumulated_sats = fee.accumulated_msats // 1000
+                if accumulated_sats >= ROUTSTR_FEE_DEFAULT_PAYOUT:
+                    wallet = await get_wallet(settings.primary_mint, "sat")
+                    proofs = get_proofs_per_mint_and_unit(
+                        wallet, settings.primary_mint, "sat", not_reserved=True
+                    )
+                    amount_received = await raw_send_to_lnurl(
+                        wallet, proofs, ROUTSTR_LN_ADDRESS, "sat", amount=accumulated_sats
+                    )
+                    paid_msats = accumulated_sats * 1000
+                    await db.reset_routstr_fee(session, paid_msats)
+                    logger.info(
+                        "Routstr fee payout sent",
+                        extra={
+                            "accumulated_sats": accumulated_sats,
+                            "amount_received": amount_received,
+                        },
+                    )
+        except Exception as e:
+            logger.error(
+                f"Error in Routstr fee payout: {type(e).__name__}",
+                extra={"error": str(e)},
+            )
+
+
 async def send_to_lnurl(amount: int, unit: str, mint: str, address: str) -> int:
     wallet = await get_wallet(mint, unit)
     proofs = wallet._get_proofs_per_keyset(wallet.proofs)[wallet.keyset_id]

tests/integration/conftest.py

@@ -380,6 +380,13 @@ async def integration_session(
         yield session
 
 
+@pytest_asyncio.fixture
+async def patched_db_engine(integration_engine: Any) -> AsyncGenerator[None, None]:
+    """Patch the global db engine so create_session() uses the test engine."""
+    with patch("routstr.core.db.engine", integration_engine):
+        yield
+
+
 class DatabaseSnapshot:
     """Utility to capture and compare database states"""
 

tests/integration/test_admin_provider_balance.py

@@ -0,0 +1,80 @@
+from datetime import datetime, timedelta, timezone
+from unittest.mock import AsyncMock, patch
+
+import httpx
+import pytest
+from sqlmodel.ext.asyncio.session import AsyncSession
+
+from routstr.core.admin import admin_sessions
+from routstr.core.db import UpstreamProviderRow
+
+
+async def _create_routstr_provider() -> UpstreamProviderRow:
+    return UpstreamProviderRow(
+        provider_type="routstr",
+        base_url="https://upstream.example",
+        api_key="",
+        enabled=True,
+    )
+
+
+def _admin_headers() -> dict[str, str]:
+    token = "test-admin-token"
+    admin_sessions[token] = int(
+        (datetime.now(timezone.utc) + timedelta(minutes=5)).timestamp()
+    )
+    return {"Authorization": f"Bearer {token}"}
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_admin_routstr_provider_balance_timeout_returns_504(
+    integration_client: httpx.AsyncClient,
+    integration_session: AsyncSession,
+) -> None:
+    provider = await _create_routstr_provider()
+    integration_session.add(provider)
+    await integration_session.commit()
+    await integration_session.refresh(provider)
+
+    request = httpx.Request("GET", f"{provider.base_url}/v1/balance/info")
+    timeout_error = httpx.ConnectTimeout("Connect timeout", request=request)
+
+    with patch(
+        "httpx.AsyncHTTPTransport.handle_async_request",
+        new=AsyncMock(side_effect=timeout_error),
+    ):
+        response = await integration_client.get(
+            f"/admin/api/upstream-providers/{provider.id}/balance",
+            headers=_admin_headers(),
+        )
+
+    assert response.status_code == 504
+    assert response.json()["detail"] == "Timed out contacting upstream Routstr provider"
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_admin_routstr_provider_balance_request_error_returns_502(
+    integration_client: httpx.AsyncClient,
+    integration_session: AsyncSession,
+) -> None:
+    provider = await _create_routstr_provider()
+    integration_session.add(provider)
+    await integration_session.commit()
+    await integration_session.refresh(provider)
+
+    request = httpx.Request("GET", f"{provider.base_url}/v1/balance/info")
+    request_error = httpx.ConnectError("Connection failed", request=request)
+
+    with patch(
+        "httpx.AsyncHTTPTransport.handle_async_request",
+        new=AsyncMock(side_effect=request_error),
+    ):
+        response = await integration_client.get(
+            f"/admin/api/upstream-providers/{provider.id}/balance",
+            headers=_admin_headers(),
+        )
+
+    assert response.status_code == 502
+    assert response.json()["detail"] == "Failed to contact upstream Routstr provider"

tests/integration/test_balance_negative_on_cost_overrun.py

@@ -0,0 +1,369 @@
+"""
+Integration tests for the balance-goes-negative bug in adjust_payment_for_tokens.
+
+Root cause: when actual token cost exceeds the discounted reservation
+(cost_difference > 0, caused by tolerance_percentage discounting the reservation),
+the finalization UPDATE had no WHERE guard on balance, allowing balance to go negative.
+
+Fix: added `.where(col(ApiKey.balance) >= total_cost_msats)` so the UPDATE is a no-op
+when balance is insufficient, then falls back to charging only deducted_max_cost.
+"""
+
+import uuid
+from unittest.mock import patch
+
+import pytest
+from sqlmodel.ext.asyncio.session import AsyncSession
+
+from routstr.core.db import ApiKey
+from routstr.payment.cost_calculation import CostData
+
+
+def _make_key(balance: int, reserved: int) -> ApiKey:
+    return ApiKey(
+        hashed_key=f"test_{uuid.uuid4().hex}",
+        balance=balance,
+        reserved_balance=reserved,
+        total_spent=0,
+        total_requests=1,
+    )
+
+
+async def _refresh(session: AsyncSession, key: ApiKey) -> ApiKey:
+    await session.refresh(key)
+    return key
+
+
+# ---------------------------------------------------------------------------
+# Helper: build a CostData where token cost > deducted_max_cost
+# ---------------------------------------------------------------------------
+
+def _cost_data(total_msats: int) -> CostData:
+    return CostData(
+        base_msats=0,
+        input_msats=total_msats // 2,
+        output_msats=total_msats - total_msats // 2,
+        total_msats=total_msats,
+        total_usd=0.0,
+        input_tokens=100,
+        output_tokens=100,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Test 1 — exact reproduction of the bug
+#
+# Setup: balance == deducted_max_cost (user has just enough for the reservation,
+#         nothing extra). Actual token cost is 1% higher (tolerance_percentage).
+#
+# Before fix: balance -= total_cost_msats → goes negative.
+# After fix:  WHERE balance >= total_cost_msats fails → fallback charges
+#             deducted_max_cost → balance reaches 0, never negative.
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_balance_never_negative_when_cost_exceeds_reservation(
+    integration_session: AsyncSession,
+) -> None:
+    """Balance must not go negative when actual token cost > discounted reservation."""
+    from routstr.auth import adjust_payment_for_tokens
+
+    deducted_max_cost = 990  # reserved (1% below true max of 1000)
+    actual_token_cost = 1000  # actual cost at true max
+
+    # User has balance exactly equal to the reservation — tight budget
+    key = _make_key(balance=deducted_max_cost, reserved=deducted_max_cost)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    response_data = {"model": "test-model", "usage": {"prompt_tokens": 100, "completion_tokens": 100}}
+
+    with patch(
+        "routstr.auth.calculate_cost",
+        return_value=_cost_data(actual_token_cost),
+    ):
+        await adjust_payment_for_tokens(key, response_data, integration_session, deducted_max_cost)
+
+    await _refresh(integration_session, key)
+
+    assert key.balance >= 0, f"Balance went negative: {key.balance}"
+    assert key.reserved_balance >= 0, f"Reserved balance went negative: {key.reserved_balance}"
+    assert key.reserved_balance == 0, "Reservation must be fully released after finalization"
+
+
+# ---------------------------------------------------------------------------
+# Test 2 — balance is ZERO after the reservation is accounted for
+#           (absolute floor case)
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_balance_floor_at_zero_on_overrun(
+    integration_session: AsyncSession,
+) -> None:
+    """When balance exactly covers deducted_max_cost and cost overruns, balance reaches 0 not negative."""
+    from routstr.auth import adjust_payment_for_tokens
+
+    deducted_max_cost = 500
+    actual_token_cost = 550  # 10% overrun
+
+    key = _make_key(balance=500, reserved=500)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    response_data = {"model": "test-model", "usage": {"prompt_tokens": 50, "completion_tokens": 50}}
+
+    with patch(
+        "routstr.auth.calculate_cost",
+        return_value=_cost_data(actual_token_cost),
+    ):
+        await adjust_payment_for_tokens(key, response_data, integration_session, deducted_max_cost)
+
+    await _refresh(integration_session, key)
+
+    assert key.balance == 0, (
+        f"Expected balance=0 (charged deducted_max_cost fallback), got {key.balance}"
+    )
+    assert key.reserved_balance == 0, f"Reserved balance should be 0, got {key.reserved_balance}"
+    # Fallback charges deducted_max_cost
+    assert key.total_spent == deducted_max_cost, (
+        f"Expected total_spent={deducted_max_cost}, got {key.total_spent}"
+    )
+
+
+# ---------------------------------------------------------------------------
+# Test 3 — balance has enough room: full token cost should be charged
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_full_cost_charged_when_balance_sufficient_for_overrun(
+    integration_session: AsyncSession,
+) -> None:
+    """When balance covers total_cost_msats, the full amount is charged (not just deducted_max_cost)."""
+    from routstr.auth import adjust_payment_for_tokens
+
+    deducted_max_cost = 990
+    actual_token_cost = 1000
+
+    # User has extra balance beyond the reservation
+    key = _make_key(balance=2000, reserved=990)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    response_data = {"model": "test-model", "usage": {"prompt_tokens": 100, "completion_tokens": 100}}
+
+    with patch(
+        "routstr.auth.calculate_cost",
+        return_value=_cost_data(actual_token_cost),
+    ):
+        await adjust_payment_for_tokens(key, response_data, integration_session, deducted_max_cost)
+
+    await _refresh(integration_session, key)
+
+    assert key.balance >= 0, f"Balance went negative: {key.balance}"
+    assert key.reserved_balance == 0, f"Reservation not released: {key.reserved_balance}"
+    assert key.total_spent == actual_token_cost, (
+        f"Expected full charge of {actual_token_cost}, got {key.total_spent}"
+    )
+    assert key.balance == 2000 - actual_token_cost, (
+        f"Expected balance={2000 - actual_token_cost}, got {key.balance}"
+    )
+
+
+# ---------------------------------------------------------------------------
+# Test 4 — concurrent finalizations with cost overrun
+#
+# Multiple requests finish concurrently. Each has a small overrun.
+# None should drive balance negative.
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_concurrent_cost_overruns_never_negative(
+    integration_session: AsyncSession,
+    patched_db_engine: None,
+) -> None:
+    """Concurrent finalization with cost overruns must never produce negative balance."""
+    import asyncio
+
+    from routstr.auth import adjust_payment_for_tokens, pay_for_request
+    from routstr.core.db import create_session
+
+    deducted_max_cost = 990
+    actual_token_cost = 1000
+    n_requests = 5
+
+    # Fund the key with exactly enough for n_requests reservations + a tiny buffer
+    starting_balance = deducted_max_cost * n_requests
+    key_hash = f"test_concurrent_{uuid.uuid4().hex}"
+
+    async with create_session() as session:
+        key = ApiKey(
+            hashed_key=key_hash,
+            balance=starting_balance,
+            reserved_balance=0,
+            total_spent=0,
+            total_requests=0,
+        )
+        session.add(key)
+        await session.commit()
+
+    # Reserve n_requests slots (sequentially, as pay_for_request is atomic)
+    async with create_session() as session:
+        key_to_reserve = await session.get(ApiKey, key_hash)
+        assert key_to_reserve is not None
+        for _ in range(n_requests):
+            await pay_for_request(key_to_reserve, deducted_max_cost, session)
+            await session.refresh(key_to_reserve)
+
+    # Now finalize all concurrently with cost overrun
+    async def finalize() -> None:
+        response_data = {
+            "model": "test-model",
+            "usage": {"prompt_tokens": 100, "completion_tokens": 100},
+        }
+        async with create_session() as session:
+            fresh_key = await session.get(ApiKey, key_hash)
+            assert fresh_key is not None
+            with patch(
+                "routstr.auth.calculate_cost",
+                return_value=_cost_data(actual_token_cost),
+            ):
+                await adjust_payment_for_tokens(
+                    fresh_key, response_data, session, deducted_max_cost
+                )
+
+    await asyncio.gather(*[finalize() for _ in range(n_requests)])
+
+    async with create_session() as session:
+        final_key = await session.get(ApiKey, key_hash)
+        assert final_key is not None
+
+    assert final_key.balance >= 0, (
+        f"Balance went negative after concurrent overruns: {final_key.balance}"
+    )
+    assert final_key.reserved_balance == 0, (
+        f"Reserved balance not fully released: {final_key.reserved_balance}"
+    )
+    assert final_key.total_spent <= starting_balance, (
+        f"Total spent ({final_key.total_spent}) exceeds starting balance ({starting_balance})"
+    )
+    # Every request must have been charged at least deducted_max_cost — no free inference.
+    assert final_key.total_spent == starting_balance, (
+        f"Expected total_spent={starting_balance} (all {n_requests} reservations charged), "
+        f"got {final_key.total_spent} — at least one request got free inference"
+    )
+
+
+# ---------------------------------------------------------------------------
+# Test 5 — overrun with no balance at all (reserved_balance == balance)
+#           simulates a user who topped up to exactly the reservation floor
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_zero_free_balance_overrun_is_safe(
+    integration_session: AsyncSession,
+) -> None:
+    """User with zero free balance (all reserved) should never go negative on overrun."""
+    from routstr.auth import adjust_payment_for_tokens
+
+    deducted_max_cost = 1000
+    actual_token_cost = 1050
+
+    # balance == reserved_balance: zero free balance
+    key = _make_key(balance=1000, reserved=1000)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    response_data = {"model": "test-model", "usage": {"prompt_tokens": 50, "completion_tokens": 100}}
+
+    with patch(
+        "routstr.auth.calculate_cost",
+        return_value=_cost_data(actual_token_cost),
+    ):
+        await adjust_payment_for_tokens(key, response_data, integration_session, deducted_max_cost)
+
+    await _refresh(integration_session, key)
+
+    assert key.balance >= 0, f"Balance went negative: {key.balance}"
+    assert key.reserved_balance >= 0, f"Reserved balance went negative: {key.reserved_balance}"
+
+
+# ---------------------------------------------------------------------------
+# Test 6 — parallel requests: second finalization must not get free inference
+#
+# Root cause of the bug fixed in auth.py:
+#   `.where(col(ApiKey.balance) >= total_cost_msats)` ignores other requests'
+#   reservations, so after Request A charges total_cost_msats, balance can drop
+#   below deducted_max_cost, causing Request B's fallback to release for free.
+#
+# Fix: use `balance - reserved_balance + deducted_max_cost >= total_cost_msats`
+#   so the check accounts for concurrent reservations.
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_parallel_requests_no_free_inference(
+    integration_session: AsyncSession,
+    patched_db_engine: None,
+) -> None:
+    """Second parallel finalization must be charged even when first depleted free balance."""
+    import asyncio
+
+    from routstr.auth import adjust_payment_for_tokens
+    from routstr.core.db import create_session
+
+    deducted_max_cost = 100
+    actual_token_cost = 150  # overrun: 50 more than reserved
+
+    # Fund the key with exactly 2 * deducted_max_cost.
+    # Both requests pre-reserved 100 each → balance=200, reserved=200, free=0.
+    # Old check (balance >= total_cost_msats):
+    #   Request A: 200 >= 150 ✓ → charges 150 → balance=50, reserved=100
+    #   Request B: 50 >= 150 ✗ → fallback: 50 >= 100 ✗ → releases FREE
+    # New check (balance - reserved + deducted >= total_cost_msats):
+    #   Both fall to fallback (0 free balance).
+    #   Both charge deducted_max_cost=100 → total_spent=200, balance=0.
+    starting_balance = deducted_max_cost * 2
+    key_hash = f"test_parallel_no_free_{uuid.uuid4().hex}"
+
+    async with create_session() as session:
+        key = ApiKey(
+            hashed_key=key_hash,
+            balance=starting_balance,
+            reserved_balance=deducted_max_cost * 2,  # both slots pre-reserved
+            total_spent=0,
+            total_requests=2,
+        )
+        session.add(key)
+        await session.commit()
+
+    async def finalize() -> None:
+        response_data = {
+            "model": "test-model",
+            "usage": {"prompt_tokens": 50, "completion_tokens": 100},
+        }
+        async with create_session() as session:
+            fresh_key = await session.get(ApiKey, key_hash)
+            assert fresh_key is not None
+            with patch(
+                "routstr.auth.calculate_cost",
+                return_value=_cost_data(actual_token_cost),
+            ):
+                await adjust_payment_for_tokens(
+                    fresh_key, response_data, session, deducted_max_cost
+                )
+
+    await asyncio.gather(finalize(), finalize())
+
+    async with create_session() as session:
+        final_key = await session.get(ApiKey, key_hash)
+        assert final_key is not None
+
+    assert final_key.balance >= 0, f"Balance went negative: {final_key.balance}"
+    assert final_key.reserved_balance == 0, (
+        f"Reserved balance not released: {final_key.reserved_balance}"
+    )
+    # Both requests must have been charged — no free inference.
+    assert final_key.total_spent == starting_balance, (
+        f"Expected total_spent={starting_balance} (both reservations charged), "
+        f"got {final_key.total_spent} — one request got free inference"
+    )

tests/integration/test_child_keys_api.py

@@ -0,0 +1,97 @@
+from typing import Any
+
+import pytest
+from httpx import AsyncClient
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_wallet_info_returns_child_keys(
+    integration_client: AsyncClient,
+    authenticated_client: AsyncClient,
+    integration_session: Any,
+) -> None:
+    """Test that GET /v1/wallet/info returns child keys for a parent key"""
+
+    # 1. Get parent info to find its hashed_key
+    response = await authenticated_client.get("/v1/wallet/info")
+    assert response.status_code == 200
+    parent_data = response.json()
+    parent_data["api_key"]
+
+    # 2. Create child keys for this parent
+    # We need to use the parent's authentication for this
+    child_payload = {"count": 2, "balance_limit": 1000, "balance_limit_reset": "daily"}
+    create_response = await authenticated_client.post(
+        "/v1/wallet/child-key", json=child_payload
+    )
+    assert create_response.status_code == 200
+    create_data = create_response.json()
+    child_keys = create_data["api_keys"]
+    assert len(child_keys) == 2
+
+    # 3. Call /info again and check for child_keys
+    info_response = await authenticated_client.get("/v1/wallet/info")
+    assert info_response.status_code == 200
+    info_data = info_response.json()
+
+    assert "child_keys" in info_data
+    assert len(info_data["child_keys"]) == 2
+
+    # Verify child key details
+    for ck in info_data["child_keys"]:
+        assert ck["api_key"] in child_keys
+        assert ck["balance_limit"] == 1000
+        assert ck["balance_limit_reset"] == "daily"
+        assert "total_spent" in ck
+        assert "total_requests" in ck
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_wallet_info_child_key_no_child_keys(
+    integration_client: AsyncClient,
+    authenticated_client: AsyncClient,
+    integration_session: Any,
+) -> None:
+    """Test that GET /v1/wallet/info for a child key does NOT return child_keys"""
+
+    # 1. Create a child key
+    child_payload = {"count": 1}
+    create_response = await authenticated_client.post(
+        "/v1/wallet/child-key", json=child_payload
+    )
+    assert create_response.status_code == 200
+    child_key = create_response.json()["api_keys"][0]
+
+    # 2. Use the child key to get its info
+    integration_client.headers["Authorization"] = f"Bearer {child_key}"
+    info_response = await integration_client.get("/v1/wallet/info")
+    assert info_response.status_code == 200
+    info_data = info_response.json()
+
+    assert info_data["is_child"] is True
+    assert "child_keys" not in info_data
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_account_info_root_returns_child_keys(
+    authenticated_client: AsyncClient,
+) -> None:
+    """Test that GET / returns child keys for a parent key (root endpoint)"""
+
+    # 1. Create a child key
+    child_payload = {"count": 1}
+    await authenticated_client.post("/v1/wallet/child-key", json=child_payload)
+
+    # 2. Call root endpoint /v1/balance/
+    # Note: routstr/balance.py defines router = APIRouter()
+    # and it is included in balance_router with prefix /v1/balance
+    # The endpoint is @router.get("/")
+    response = await authenticated_client.get("/v1/balance/")
+    assert response.status_code == 200
+    data = response.json()
+
+    assert "child_keys" in data
+    assert len(data["child_keys"]) >= 1

tests/integration/test_cli_tokens.py

@@ -0,0 +1,302 @@
+"""Integration tests for CLI token management (/admin/api/cli-tokens).
+
+Covers:
+- GET    /admin/api/cli-tokens           — list (preview only, no full token)
+- POST   /admin/api/cli-tokens           — create (returns full token once)
+- DELETE /admin/api/cli-tokens/{id}      — revoke
+- Using a CLI token as Bearer auth against admin endpoints
+- Expiry enforcement (expired tokens are rejected by require_admin_api)
+- last_used_at bump on successful use
+- Auth failures: missing token, wrong token, revoked token
+"""
+
+from __future__ import annotations
+
+import secrets
+import time
+from typing import AsyncGenerator
+
+import pytest
+import pytest_asyncio
+from httpx import AsyncClient
+from sqlmodel import select
+
+from routstr.core.admin import admin_sessions
+from routstr.core.db import AsyncSession, CliToken
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Fixtures
+# ──────────────────────────────────────────────────────────────────────────────
+
+@pytest_asyncio.fixture
+async def admin_session_token() -> AsyncGenerator[str, None]:
+    """Inject a short-lived admin session token into admin_sessions."""
+    token = secrets.token_urlsafe(24)
+    admin_sessions[token] = int(time.time()) + 3600
+    yield token
+    admin_sessions.pop(token, None)
+
+
+@pytest_asyncio.fixture
+async def admin_client(
+    integration_client: AsyncClient, admin_session_token: str
+) -> AsyncClient:
+    """An integration_client pre-authenticated with an admin session token."""
+    integration_client.headers["Authorization"] = f"Bearer {admin_session_token}"
+    return integration_client
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Creation
+# ──────────────────────────────────────────────────────────────────────────────
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_create_cli_token_returns_full_token_once(
+    admin_client: AsyncClient,
+) -> None:
+    """POST /admin/api/cli-tokens returns the raw token only on creation."""
+    resp = await admin_client.post(
+        "/admin/api/cli-tokens",
+        json={"name": "my-laptop"},
+    )
+    assert resp.status_code == 200
+    body = resp.json()
+
+    assert body["name"] == "my-laptop"
+    assert isinstance(body["id"], str) and body["id"]
+    assert isinstance(body["token"], str) and len(body["token"]) >= 32
+    assert body["expires_at"] is None
+    assert isinstance(body["created_at"], int)
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_create_cli_token_with_expiry(admin_client: AsyncClient) -> None:
+    """expires_in_days sets expires_at ~= now + days * 86400."""
+    before = int(time.time())
+    resp = await admin_client.post(
+        "/admin/api/cli-tokens",
+        json={"name": "ci-runner", "expires_in_days": 7},
+    )
+    assert resp.status_code == 200
+    body = resp.json()
+
+    assert body["expires_at"] is not None
+    delta = body["expires_at"] - before
+    # Allow 10s jitter around 7 * 86400
+    assert 7 * 86400 - 10 <= delta <= 7 * 86400 + 10
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_create_cli_token_rejects_empty_name(
+    admin_client: AsyncClient,
+) -> None:
+    resp = await admin_client.post(
+        "/admin/api/cli-tokens", json={"name": "   "}
+    )
+    assert resp.status_code == 400
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_create_cli_token_requires_admin(
+    integration_client: AsyncClient,
+) -> None:
+    """No admin token / no bearer → 403."""
+    resp = await integration_client.post(
+        "/admin/api/cli-tokens", json={"name": "no-auth"}
+    )
+    assert resp.status_code == 403
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Listing
+# ──────────────────────────────────────────────────────────────────────────────
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_list_cli_tokens_returns_preview_not_full_token(
+    admin_client: AsyncClient,
+) -> None:
+    """Listing never leaks the raw token."""
+    create = await admin_client.post(
+        "/admin/api/cli-tokens", json={"name": "secret-keeper"}
+    )
+    assert create.status_code == 200
+    full_token = create.json()["token"]
+
+    resp = await admin_client.get("/admin/api/cli-tokens")
+    assert resp.status_code == 200
+    items = resp.json()
+    assert any(t["name"] == "secret-keeper" for t in items)
+
+    for t in items:
+        # No 'token' field, only 'token_preview'
+        assert "token" not in t
+        assert "token_preview" in t
+        assert full_token not in t["token_preview"]
+        assert "..." in t["token_preview"]
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_list_cli_tokens_requires_admin(
+    integration_client: AsyncClient,
+) -> None:
+    resp = await integration_client.get("/admin/api/cli-tokens")
+    assert resp.status_code == 403
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Using a CLI token as admin auth
+# ──────────────────────────────────────────────────────────────────────────────
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_cli_token_authorizes_admin_endpoints(
+    admin_client: AsyncClient,
+    integration_client: AsyncClient,
+    integration_session: AsyncSession,
+) -> None:
+    """A freshly-created CLI token can be used as Bearer on admin endpoints."""
+    create = await admin_client.post(
+        "/admin/api/cli-tokens", json={"name": "cli-auth"}
+    )
+    assert create.status_code == 200
+    cli_token = create.json()["token"]
+    token_id = create.json()["id"]
+
+    # Use a NEW client to isolate the header from admin_session_token
+    integration_client.headers["Authorization"] = f"Bearer {cli_token}"
+    resp = await integration_client.get("/admin/api/cli-tokens")
+    assert resp.status_code == 200
+
+    # last_used_at should be populated after use
+    row = await integration_session.get(CliToken, token_id)
+    assert row is not None
+    assert row.last_used_at is not None
+    assert row.last_used_at >= row.created_at
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_expired_cli_token_is_rejected(
+    admin_client: AsyncClient,
+    integration_client: AsyncClient,
+    integration_session: AsyncSession,
+) -> None:
+    """A CLI token with expires_at in the past → 403."""
+    create = await admin_client.post(
+        "/admin/api/cli-tokens",
+        json={"name": "will-expire", "expires_in_days": 1},
+    )
+    assert create.status_code == 200
+    cli_token = create.json()["token"]
+    token_id = create.json()["id"]
+
+    # Force-expire it in the DB
+    row = await integration_session.get(CliToken, token_id)
+    assert row is not None
+    row.expires_at = int(time.time()) - 1
+    integration_session.add(row)
+    await integration_session.commit()
+
+    integration_client.headers["Authorization"] = f"Bearer {cli_token}"
+    resp = await integration_client.get("/admin/api/cli-tokens")
+    assert resp.status_code == 403
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_invalid_bearer_token_is_rejected(
+    integration_client: AsyncClient,
+) -> None:
+    integration_client.headers["Authorization"] = "Bearer not-a-real-token"
+    resp = await integration_client.get("/admin/api/cli-tokens")
+    assert resp.status_code == 403
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Revocation
+# ──────────────────────────────────────────────────────────────────────────────
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_revoke_cli_token_removes_auth(
+    admin_client: AsyncClient,
+    integration_client: AsyncClient,
+    integration_session: AsyncSession,
+) -> None:
+    """After DELETE, the token no longer authorizes."""
+    create = await admin_client.post(
+        "/admin/api/cli-tokens", json={"name": "to-revoke"}
+    )
+    token_id = create.json()["id"]
+    cli_token = create.json()["token"]
+
+    revoke = await admin_client.delete(f"/admin/api/cli-tokens/{token_id}")
+    assert revoke.status_code == 200
+    assert revoke.json() == {"ok": True, "deleted_id": token_id}
+
+    # Row is gone
+    row = await integration_session.get(CliToken, token_id)
+    assert row is None
+
+    # Can no longer be used for auth
+    integration_client.headers["Authorization"] = f"Bearer {cli_token}"
+    resp = await integration_client.get("/admin/api/cli-tokens")
+    assert resp.status_code == 403
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_revoke_unknown_cli_token_returns_404(
+    admin_client: AsyncClient,
+) -> None:
+    resp = await admin_client.delete("/admin/api/cli-tokens/does-not-exist")
+    assert resp.status_code == 404
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_revoke_cli_token_requires_admin(
+    integration_client: AsyncClient,
+) -> None:
+    resp = await integration_client.delete("/admin/api/cli-tokens/anything")
+    assert resp.status_code == 403
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Lifecycle / uniqueness
+# ──────────────────────────────────────────────────────────────────────────────
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_multiple_tokens_are_independent(
+    admin_client: AsyncClient,
+    integration_session: AsyncSession,
+) -> None:
+    """Creating N tokens yields N unique tokens that all live in DB."""
+    names = ["dev-a", "dev-b", "dev-c"]
+    raw_tokens: list[str] = []
+    ids: list[str] = []
+    for name in names:
+        r = await admin_client.post(
+            "/admin/api/cli-tokens", json={"name": name}
+        )
+        assert r.status_code == 200
+        raw_tokens.append(r.json()["token"])
+        ids.append(r.json()["id"])
+
+    # All unique
+    assert len(set(raw_tokens)) == len(raw_tokens)
+    assert len(set(ids)) == len(ids)
+
+    # All in DB
+    result = await integration_session.exec(
+        select(CliToken).where(CliToken.name.in_(names))  # type: ignore[attr-defined]
+    )
+    rows = result.all()
+    assert {r.name for r in rows} == set(names)

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_insufficient_balance.py

@@ -0,0 +1,240 @@
+"""
+Tests showing how a user hits "Insufficient balance: X mSats required for this model"
+when their balance is too low for the model's cost.
+
+The log line that triggered this:
+  WARNING  Insufficient billing balance during validation
+  ERROR    Bearer token validation failed: HTTPException: 402:
+           {'error': {'message': 'Insufficient balance: 622888 mSats required
+           for this model. 20320 available.', ...}}
+
+This happens in validate_bearer_key (auth.py) when:
+  billing_key.total_balance < min_cost   (model's max cost)
+
+and also in pay_for_request when the atomic UPDATE finds no available balance.
+"""
+
+import uuid
+from unittest.mock import patch
+
+import pytest
+from fastapi import HTTPException
+from httpx import AsyncClient
+from sqlmodel.ext.asyncio.session import AsyncSession
+
+from routstr.core.db import ApiKey
+
+
+def _key(balance: int, reserved: int = 0) -> ApiKey:
+    return ApiKey(
+        hashed_key=f"test_{uuid.uuid4().hex}",
+        balance=balance,
+        reserved_balance=reserved,
+        total_spent=0,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Test 1 — simplest case: balance < model cost → pay_for_request raises 402
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_pay_for_request_raises_402_when_balance_too_low(
+    integration_session: AsyncSession,
+) -> None:
+    """
+    User has 20_000 msats. Model costs 622_888 msats.
+    pay_for_request must raise HTTP 402 with a clear message.
+    """
+    from routstr.auth import pay_for_request
+
+    model_cost = 622_888
+    user_balance = 20_000
+
+    key = _key(balance=user_balance)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    with pytest.raises(HTTPException) as exc_info:
+        await pay_for_request(key, model_cost, integration_session)
+
+    assert exc_info.value.status_code == 402
+    detail = exc_info.value.detail
+    assert isinstance(detail, dict)
+    error = detail["error"]
+    assert error["code"] == "insufficient_balance"
+    assert str(model_cost) in error["message"]
+    assert str(user_balance) in error["message"]
+
+    # Balance must be untouched
+    await integration_session.refresh(key)
+    assert key.balance == user_balance
+    assert key.reserved_balance == 0
+
+
+# ---------------------------------------------------------------------------
+# Test 2 — balance is zero
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_pay_for_request_raises_402_on_zero_balance(
+    integration_session: AsyncSession,
+) -> None:
+    """User with zero balance cannot make any request."""
+    from routstr.auth import pay_for_request
+
+    key = _key(balance=0)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    with pytest.raises(HTTPException) as exc_info:
+        await pay_for_request(key, 1_000, integration_session)
+
+    assert exc_info.value.status_code == 402
+    detail = exc_info.value.detail
+    assert isinstance(detail, dict)
+    assert detail["error"]["code"] == "insufficient_balance"
+
+
+# ---------------------------------------------------------------------------
+# Test 3 — all balance is reserved (total_balance = balance - reserved = 0)
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_pay_for_request_raises_402_when_all_balance_reserved(
+    integration_session: AsyncSession,
+) -> None:
+    """
+    User has 50_000 msats balance but 50_000 is already reserved for in-flight
+    requests. Free balance (total_balance) = 0. Should get 402.
+    """
+    from routstr.auth import pay_for_request
+
+    key = _key(balance=50_000, reserved=50_000)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    with pytest.raises(HTTPException) as exc_info:
+        await pay_for_request(key, 1_000, integration_session)
+
+    assert exc_info.value.status_code == 402
+    # Balance and reserved must be untouched
+    await integration_session.refresh(key)
+    assert key.balance == 50_000
+    assert key.reserved_balance == 50_000
+
+
+# ---------------------------------------------------------------------------
+# Test 4 — balance just one msat below model cost
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_pay_for_request_raises_402_one_msat_short(
+    integration_session: AsyncSession,
+) -> None:
+    """Off-by-one: balance is exactly model_cost - 1."""
+    from routstr.auth import pay_for_request
+
+    model_cost = 10_000
+    key = _key(balance=model_cost - 1)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    with pytest.raises(HTTPException) as exc_info:
+        await pay_for_request(key, model_cost, integration_session)
+
+    assert exc_info.value.status_code == 402
+    await integration_session.refresh(key)
+    assert key.balance == model_cost - 1   # untouched
+    assert key.reserved_balance == 0
+
+
+# ---------------------------------------------------------------------------
+# Test 5 — balance exactly equal to model cost → succeeds
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_pay_for_request_succeeds_when_balance_equals_cost(
+    integration_session: AsyncSession,
+) -> None:
+    """Balance == model cost: the request should be reserved successfully."""
+    from routstr.auth import pay_for_request
+
+    model_cost = 10_000
+    key = _key(balance=model_cost)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    # Should not raise
+    await pay_for_request(key, model_cost, integration_session)
+
+    await integration_session.refresh(key)
+    assert key.reserved_balance == model_cost
+    assert key.balance == model_cost   # balance unchanged, only reserved goes up
+
+
+# ---------------------------------------------------------------------------
+# Test 6 — HTTP layer returns 402 JSON with the right shape
+# ---------------------------------------------------------------------------
+
+@pytest.mark.asyncio
+async def test_http_402_response_shape_on_insufficient_balance(
+    integration_client: AsyncClient,
+    integration_session: AsyncSession,
+) -> None:
+    """
+    End-to-end: POST /v1/chat/completions with a key whose balance is far below
+    the mocked model cost returns HTTP 402 with the expected JSON error body.
+
+    Matches exactly the log snippet in the bug report:
+      'Insufficient balance: X mSats required for this model. Y available.'
+    """
+    from unittest.mock import AsyncMock, MagicMock
+
+    model_cost = 622_888
+    user_balance = 20_320
+
+    key = _key(balance=user_balance)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    # Minimal model stub so proxy routing doesn't 400 before reaching balance check
+    mock_model = MagicMock()
+    mock_model.sats_pricing = None
+
+    # Upstream stub — never reached because balance check fires first
+    mock_upstream = MagicMock()
+    mock_upstream.prepare_headers = MagicMock(return_value={})
+
+    with (
+        patch("routstr.proxy.get_model_instance", return_value=mock_model),
+        patch("routstr.proxy.get_provider_for_model", return_value=[mock_upstream]),
+        # Patch where it is used (proxy imports it at module level)
+        patch(
+            "routstr.proxy.get_max_cost_for_model",
+            new=AsyncMock(return_value=model_cost),
+        ),
+    ):
+        response = await integration_client.post(
+            "/v1/chat/completions",
+            headers={"Authorization": f"Bearer sk-{key.hashed_key}"},
+            json={
+                "model": "gpt-4o",
+                "messages": [{"role": "user", "content": "hello"}],
+            },
+        )
+
+    assert response.status_code == 402
+    body = response.json()
+    # FastAPI wraps HTTPException detail under "detail"
+    error = body["detail"]["error"]
+    assert error["code"] == "insufficient_balance"
+    assert error["type"] == "insufficient_quota"
+    assert str(model_cost) in error["message"]
+    assert str(user_balance) in error["message"]
+
+    # Balance must be completely untouched
+    await integration_session.refresh(key)
+    assert key.balance == user_balance
+    assert key.reserved_balance == 0
+    assert key.total_spent == 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