fix provider model pricing

Allow multiple providers per base URL

docs/api/endpoints.md

@@ -434,6 +434,42 @@ Authorization: Bearer sk-...
 }
 ```
 
+### Create Child Key
+
+Creates one or more child API keys that share the parent's balance. Each child key creation costs a fixed amount (configurable).
+
+```http
+POST /v1/balance/child-key
+Authorization: Bearer sk-...
+```
+
+**Request Body:**
+
+```json
+{
+  "count": 1
+}
+```
+
+**Parameters:**
+
+| Parameter | Type | Required | Default | Description |
+|-----------|------|----------|---------|-------------|
+| `count` | integer | Yes | - | Number of child keys to create (1-50) |
+
+**Response:**
+
+```json
+{
+  "api_keys": ["sk-abc...", "sk-def..."],
+  "count": 2,
+  "cost_msats": 2000,
+  "cost_sats": 2,
+  "parent_balance": 98000,
+  "parent_balance_sats": 98
+}
+```
+
 ## Provider Discovery
 
 ## Admin Settings

examples/create_child_keys.py

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

migrations/versions/a86e5348850b_.py

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

migrations/versions/c2d3e4f5a6b7_upstream_provider_base_url_api_key_unique.py

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

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "routstr"
-version = "0.2.2"
+version = "0.3.0"
 description = "Payment proxy for your LLM endpoint using cashu and nostr."
 readme = "README.md"
 requires-python = ">=3.11"

routstr/algorithm.py

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

routstr/auth.py

@@ -286,30 +286,55 @@ async def validate_bearer_key(
     )
 
 
+async def get_billing_key(key: ApiKey, session: AsyncSession) -> ApiKey:
+    """Returns the key that should be charged for the request."""
+    if key.parent_key_hash:
+        parent = await session.get(ApiKey, key.parent_key_hash)
+        if parent:
+            # We want to keep the total_requests and total_spent on the child key
+            # but use the balance and reserved_balance of the parent.
+            # However, pay_for_request updates reserved_balance and total_requests.
+            # To stay simple, we charge the parent's balance and update parent's total_requests.
+            return parent
+        else:
+            logger.error(
+                "Parent key not found for child key",
+                extra={
+                    "child_key_hash": key.hashed_key[:8] + "...",
+                    "parent_key_hash": key.parent_key_hash[:8] + "...",
+                },
+            )
+    return key
+
+
 async def pay_for_request(
     key: ApiKey, cost_per_request: int, session: AsyncSession
 ) -> int:
     """Process payment for a request."""
 
+    billing_key = await get_billing_key(key, session)
+
     logger.info(
         "Processing payment for request",
         extra={
             "key_hash": key.hashed_key[:8] + "...",
-            "current_balance": key.balance,
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
+            "current_balance": billing_key.balance,
             "required_cost": cost_per_request,
-            "sufficient_balance": key.balance >= cost_per_request,
+            "sufficient_balance": billing_key.balance >= cost_per_request,
         },
     )
 
-    if key.total_balance < cost_per_request:
+    if billing_key.total_balance < cost_per_request:
         logger.warning(
             "Insufficient balance for request",
             extra={
                 "key_hash": key.hashed_key[:8] + "...",
-                "balance": key.balance,
-                "reserved_balance": key.reserved_balance,
+                "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                "balance": billing_key.balance,
+                "reserved_balance": billing_key.reserved_balance,
                 "required": cost_per_request,
-                "shortfall": cost_per_request - key.total_balance,
+                "shortfall": cost_per_request - billing_key.total_balance,
             },
         )
 
@@ -317,7 +342,7 @@ async def pay_for_request(
             status_code=402,
             detail={
                 "error": {
-                    "message": f"Insufficient balance: {cost_per_request} mSats required. {key.total_balance} available. (reserved: {key.reserved_balance})",
+                    "message": f"Insufficient balance: {cost_per_request} mSats required. {billing_key.total_balance} available. (reserved: {billing_key.reserved_balance})",
                     "type": "insufficient_quota",
                     "code": "insufficient_balance",
                 }
@@ -328,15 +353,16 @@ async def pay_for_request(
         "Charging base cost for request",
         extra={
             "key_hash": key.hashed_key[:8] + "...",
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
             "cost": cost_per_request,
-            "balance_before": key.balance,
+            "balance_before": billing_key.balance,
         },
     )
 
     # Charge the base cost for the request atomically to avoid race conditions
     stmt = (
         update(ApiKey)
-        .where(col(ApiKey.hashed_key) == key.hashed_key)
+        .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
         .where(col(ApiKey.balance) - col(ApiKey.reserved_balance) >= cost_per_request)
         .values(
             reserved_balance=col(ApiKey.reserved_balance) + cost_per_request,
@@ -344,6 +370,16 @@ 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
+    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)
+        )
+        await session.exec(child_stmt)  # type: ignore[call-overload]
+
     await session.commit()
 
     if result.rowcount == 0:
@@ -351,8 +387,9 @@ async def pay_for_request(
             "Concurrent request depleted balance",
             extra={
                 "key_hash": key.hashed_key[:8] + "...",
+                "billing_key_hash": billing_key.hashed_key[:8] + "...",
                 "required_cost": cost_per_request,
-                "current_balance": key.balance,
+                "current_balance": billing_key.balance,
             },
         )
 
@@ -361,23 +398,26 @@ async def pay_for_request(
             status_code=402,
             detail={
                 "error": {
-                    "message": f"Insufficient balance: {cost_per_request} mSats required. {key.balance} available.",
+                    "message": f"Insufficient balance: {cost_per_request} mSats required. {billing_key.balance} available.",
                     "type": "insufficient_quota",
                     "code": "insufficient_balance",
                 }
             },
         )
 
-    await session.refresh(key)
+    await session.refresh(billing_key)
+    if billing_key.hashed_key != key.hashed_key:
+        await session.refresh(key)
 
     logger.info(
         "Payment processed successfully",
         extra={
             "key_hash": key.hashed_key[:8] + "...",
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
             "charged_amount": cost_per_request,
-            "new_balance": key.balance,
-            "total_spent": key.total_spent,
-            "total_requests": key.total_requests,
+            "new_balance": billing_key.balance,
+            "total_spent": billing_key.total_spent,
+            "total_requests": billing_key.total_requests,
         },
     )
 
@@ -387,9 +427,11 @@ async def pay_for_request(
 async def revert_pay_for_request(
     key: ApiKey, session: AsyncSession, cost_per_request: int
 ) -> None:
+    billing_key = await get_billing_key(key, session)
+
     stmt = (
         update(ApiKey)
-        .where(col(ApiKey.hashed_key) == key.hashed_key)
+        .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
         .values(
             reserved_balance=col(ApiKey.reserved_balance) - cost_per_request,
             total_requests=col(ApiKey.total_requests) - 1,
@@ -397,27 +439,40 @@ 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
+    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)
+        )
+        await session.exec(child_stmt)  # type: ignore[call-overload]
+
     await session.commit()
     if result.rowcount == 0:
         logger.error(
             "Failed to revert payment - insufficient reserved balance",
             extra={
                 "key_hash": key.hashed_key[:8] + "...",
+                "billing_key_hash": billing_key.hashed_key[:8] + "...",
                 "cost_to_revert": cost_per_request,
-                "current_reserved_balance": key.reserved_balance,
+                "current_reserved_balance": billing_key.reserved_balance,
             },
         )
         raise HTTPException(
             status_code=402,
             detail={
                 "error": {
-                    "message": f"failed to revert request payment: {cost_per_request} mSats required. {key.balance} available.",
+                    "message": f"failed to revert request payment: {cost_per_request} mSats required. {billing_key.balance} available.",
                     "type": "payment_error",
                     "code": "payment_error",
                 }
             },
         )
-    await session.refresh(key)
+    await session.refresh(billing_key)
+    if billing_key.hashed_key != key.hashed_key:
+        await session.refresh(key)
 
 
 async def adjust_payment_for_tokens(
@@ -428,15 +483,17 @@ async def adjust_payment_for_tokens(
     This is called after the initial payment and the upstream request is complete.
     Returns cost data to be included in the response.
     """
+    billing_key = await get_billing_key(key, session)
     model = response_data.get("model", "unknown")
 
     logger.debug(
         "Starting payment adjustment for tokens",
         extra={
             "key_hash": key.hashed_key[:8] + "...",
+            "billing_key_hash": billing_key.hashed_key[:8] + "...",
             "model": model,
             "deducted_max_cost": deducted_max_cost,
-            "current_balance": key.balance,
+            "current_balance": billing_key.balance,
             "has_usage": "usage" in response_data,
         },
     )
@@ -446,8 +503,10 @@ async def adjust_payment_for_tokens(
         try:
             release_stmt = (
                 update(ApiKey)
-                .where(col(ApiKey.hashed_key) == key.hashed_key)
-                .values(reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost)
+                .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
+                .values(
+                    reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost
+                )
             )
             await session.exec(release_stmt)  # type: ignore[call-overload]
             await session.commit()
@@ -455,13 +514,18 @@ async def adjust_payment_for_tokens(
                 "Released reservation without charging (fallback)",
                 extra={
                     "key_hash": key.hashed_key[:8] + "...",
+                    "billing_key_hash": billing_key.hashed_key[:8] + "...",
                     "deducted_max_cost": deducted_max_cost,
                 },
             )
         except Exception as e:
             logger.error(
                 "Failed to release reservation in fallback",
-                extra={"error": str(e), "key_hash": key.hashed_key[:8] + "..."},
+                extra={
+                    "error": str(e),
+                    "key_hash": key.hashed_key[:8] + "...",
+                    "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                },
             )
 
     match await calculate_cost(response_data, deducted_max_cost, session):
@@ -470,6 +534,7 @@ async def adjust_payment_for_tokens(
                 "Using max cost data (no token adjustment)",
                 extra={
                     "key_hash": key.hashed_key[:8] + "...",
+                    "billing_key_hash": billing_key.hashed_key[:8] + "...",
                     "model": model,
                     "max_cost": cost.total_msats,
                 },
@@ -477,7 +542,7 @@ async def adjust_payment_for_tokens(
             # Finalize by releasing reservation and charging max cost
             finalize_stmt = (
                 update(ApiKey)
-                .where(col(ApiKey.hashed_key) == key.hashed_key)
+                .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                 .values(
                     reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost,
                     balance=col(ApiKey.balance) - cost.total_msats,
@@ -485,27 +550,41 @@ async def adjust_payment_for_tokens(
                 )
             )
             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) + cost.total_msats)
+                )
+                await session.exec(child_stmt)  # type: ignore[call-overload]
+
             await session.commit()
             if result.rowcount == 0:
                 logger.error(
                     "Failed to finalize max-cost payment - retrying reservation release",
                     extra={
                         "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
                         "deducted_max_cost": deducted_max_cost,
-                        "current_reserved_balance": key.reserved_balance,
+                        "current_reserved_balance": billing_key.reserved_balance,
                         "total_cost": cost.total_msats,
                         "model": model,
                     },
                 )
                 await release_reservation_only()
             else:
-                await session.refresh(key)
+                await session.refresh(billing_key)
+                if billing_key.hashed_key != key.hashed_key:
+                    await session.refresh(key)
                 logger.info(
                     "Max cost payment finalized",
                     extra={
                         "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
                         "charged_amount": cost.total_msats,
-                        "new_balance": key.balance,
+                        "new_balance": billing_key.balance,
                         "model": model,
                     },
                 )
@@ -521,6 +600,7 @@ async def adjust_payment_for_tokens(
                 "Calculated token-based cost",
                 extra={
                     "key_hash": key.hashed_key[:8] + "...",
+                    "billing_key_hash": billing_key.hashed_key[:8] + "...",
                     "model": model,
                     "token_cost": cost.total_msats,
                     "deducted_max_cost": deducted_max_cost,
@@ -533,11 +613,15 @@ async def adjust_payment_for_tokens(
             if cost_difference == 0:
                 logger.debug(
                     "Finalizing with exact reserved cost",
-                    extra={"key_hash": key.hashed_key[:8] + "...", "model": model},
+                    extra={
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                        "model": model,
+                    },
                 )
                 finalize_stmt = (
                     update(ApiKey)
-                    .where(col(ApiKey.hashed_key) == key.hashed_key)
+                    .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                     .values(
                         reserved_balance=col(ApiKey.reserved_balance)
                         - deducted_max_cost,
@@ -546,8 +630,20 @@ async def adjust_payment_for_tokens(
                     )
                 )
                 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)
+                    )
+                    await session.exec(child_stmt)  # type: ignore[call-overload]
+
                 await session.commit()
-                await session.refresh(key)
+                await session.refresh(billing_key)
+                if billing_key.hashed_key != key.hashed_key:
+                    await session.refresh(key)
                 return cost.dict()
 
             # this should never happen why do we handle this???
@@ -557,16 +653,17 @@ async def adjust_payment_for_tokens(
                     "Additional charge required for token usage",
                     extra={
                         "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
                         "additional_charge": cost_difference,
-                        "current_balance": key.balance,
-                        "sufficient_balance": key.balance >= cost_difference,
+                        "current_balance": billing_key.balance,
+                        "sufficient_balance": billing_key.balance >= cost_difference,
                         "model": model,
                     },
                 )
 
                 finalize_stmt = (
                     update(ApiKey)
-                    .where(col(ApiKey.hashed_key) == key.hashed_key)
+                    .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                     .values(
                         reserved_balance=col(ApiKey.reserved_balance)
                         - deducted_max_cost,
@@ -575,18 +672,31 @@ async def adjust_payment_for_tokens(
                     )
                 )
                 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)
+                    )
+                    await session.exec(child_stmt)  # type: ignore[call-overload]
+
                 await session.commit()
 
                 if result.rowcount:
                     cost.total_msats = total_cost_msats
-                    await session.refresh(key)
+                    await session.refresh(billing_key)
+                    if billing_key.hashed_key != key.hashed_key:
+                        await session.refresh(key)
 
                     logger.info(
                         "Finalized payment with additional charge",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
                             "charged_amount": total_cost_msats,
-                            "new_balance": key.balance,
+                            "new_balance": billing_key.balance,
                             "model": model,
                         },
                     )
@@ -595,6 +705,7 @@ async def adjust_payment_for_tokens(
                         "Failed to finalize additional charge - releasing reservation",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
                             "attempted_charge": total_cost_msats,
                             "model": model,
                         },
@@ -607,15 +718,16 @@ async def adjust_payment_for_tokens(
                     "Refunding excess payment",
                     extra={
                         "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
                         "refund_amount": refund,
-                        "current_balance": key.balance,
+                        "current_balance": billing_key.balance,
                         "model": model,
                     },
                 )
 
                 refund_stmt = (
                     update(ApiKey)
-                    .where(col(ApiKey.hashed_key) == key.hashed_key)
+                    .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
                     .values(
                         reserved_balance=col(ApiKey.reserved_balance)
                         - deducted_max_cost,
@@ -624,6 +736,16 @@ async def adjust_payment_for_tokens(
                     )
                 )
                 result = await session.exec(refund_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)
+                    )
+                    await session.exec(child_stmt)  # type: ignore[call-overload]
+
                 await session.commit()
 
                 if result.rowcount == 0:
@@ -631,8 +753,9 @@ async def adjust_payment_for_tokens(
                         "Failed to finalize payment - releasing reservation",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
                             "deducted_max_cost": deducted_max_cost,
-                            "current_reserved_balance": key.reserved_balance,
+                            "current_reserved_balance": billing_key.reserved_balance,
                             "total_cost": total_cost_msats,
                             "model": model,
                         },
@@ -640,14 +763,17 @@ async def adjust_payment_for_tokens(
                     await release_reservation_only()
                 else:
                     cost.total_msats = total_cost_msats
-                    await session.refresh(key)
+                    await session.refresh(billing_key)
+                    if billing_key.hashed_key != key.hashed_key:
+                        await session.refresh(key)
 
                     logger.info(
                         "Refund processed successfully",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
+                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
                             "refunded_amount": refund,
-                            "new_balance": key.balance,
+                            "new_balance": billing_key.balance,
                             "final_cost": cost.total_msats,
                             "model": model,
                         },

routstr/balance.py

@@ -32,16 +32,30 @@ async def get_key_from_header(
     )
 
 
-# TODO: remove this endpoint when frontend is updated
-@router.get("/", include_in_schema=False)
-async def account_info(key: ApiKey = Depends(get_key_from_header)) -> dict:
+async def get_balance_info(key: ApiKey, session: AsyncSession) -> dict:
+    from .auth import get_billing_key
+
+    billing_key = await get_billing_key(key, session)
     return {
         "api_key": "sk-" + key.hashed_key,
-        "balance": key.balance,
-        "reserved": key.reserved_balance,
+        "balance": billing_key.balance,
+        "reserved": billing_key.reserved_balance,
+        "is_child": key.parent_key_hash is not None,
+        "parent_key": "sk-" + key.parent_key_hash if key.parent_key_hash else None,
+        "total_requests": key.total_requests,
+        "total_spent": key.total_spent,
     }
 
 
+# TODO: remove this endpoint when frontend is updated
+@router.get("/", include_in_schema=False)
+async def account_info(
+    key: ApiKey = Depends(get_key_from_header),
+    session: AsyncSession = Depends(get_session),
+) -> dict:
+    return await get_balance_info(key, session)
+
+
 # TODO: Implement POST /v1/wallet/create endpoint
 # This endpoint should accept:
 # - cashu_token (required): The eCash token to deposit
@@ -66,12 +80,11 @@ async def create_balance(
 
 
 @router.get("/info")
-async def wallet_info(key: ApiKey = Depends(get_key_from_header)) -> dict:
-    return {
-        "api_key": "sk-" + key.hashed_key,
-        "balance": key.balance,
-        "reserved": key.reserved_balance,
-    }
+async def wallet_info(
+    key: ApiKey = Depends(get_key_from_header),
+    session: AsyncSession = Depends(get_session),
+) -> dict:
+    return await get_balance_info(key, session)
 
 
 class TopupRequest(BaseModel):
@@ -85,6 +98,10 @@ async def topup_wallet_endpoint(
     key: ApiKey = Depends(get_key_from_header),
     session: AsyncSession = Depends(get_session),
 ) -> dict[str, int]:
+    from .auth import get_billing_key
+
+    billing_key = await get_billing_key(key, session)
+
     if topup_request is not None:
         cashu_token = topup_request.cashu_token
     if cashu_token is None:
@@ -94,7 +111,7 @@ async def topup_wallet_endpoint(
     if len(cashu_token) < 10 or "cashu" not in cashu_token:
         raise HTTPException(status_code=400, detail="Invalid token format")
     try:
-        amount_msats = await credit_balance(cashu_token, key, session)
+        amount_msats = await credit_balance(cashu_token, billing_key, session)
     except ValueError as e:
         error_msg = str(e)
         if "already spent" in error_msg.lower():
@@ -155,6 +172,12 @@ async def refund_wallet_endpoint(
 
     key: ApiKey = await validate_bearer_key(bearer_value, session)
 
+    if key.parent_key_hash:
+        raise HTTPException(
+            status_code=400,
+            detail="Cannot refund child key. Please refund the parent key instead.",
+        )
+
     remaining_balance_msats: int = key.total_balance
 
     if key.refund_currency == "sat":
@@ -228,6 +251,75 @@ async def donate(token: str, ref: str | None = None) -> str:
         return "Invalid token."
 
 
+class ChildKeyRequest(BaseModel):
+    count: int
+
+
+@router.post("/child-key")
+async def create_child_key(
+    payload: ChildKeyRequest,
+    key: ApiKey = Depends(get_key_from_header),
+    session: AsyncSession = Depends(get_session),
+) -> dict:
+    """Creates one or more child API keys that use the parent's balance."""
+    # Log incoming request for debugging
+    logger.debug(f"Child key creation request: count={payload.count}")
+
+    count = payload.count
+    if count < 1 or count > 50:
+        raise HTTPException(status_code=400, detail="Count must be between 1 and 50.")
+
+    # Check if this is already a child key
+    if key.parent_key_hash:
+        raise HTTPException(
+            status_code=400,
+            detail="Cannot create a child key for another child key.",
+        )
+
+    cost_per_key = settings.child_key_cost
+    total_cost = cost_per_key * count
+
+    if key.total_balance < total_cost:
+        raise HTTPException(
+            status_code=402,
+            detail=f"Insufficient balance to create {count} child keys. {total_cost} mSats required.",
+        )
+
+    # Deduct cost from parent
+    key.balance -= total_cost
+    key.total_spent += total_cost
+    session.add(key)
+
+    # Generate new keys
+    import secrets
+
+    new_keys = []
+    for _ in range(count):
+        new_key_raw = secrets.token_hex(32)
+        new_key_hash = new_key_raw  # We use the raw key as the hash for sk- keys
+
+        child_key = ApiKey(
+            hashed_key=new_key_hash,
+            balance=0,
+            parent_key_hash=key.hashed_key,
+        )
+        session.add(child_key)
+        new_keys.append("sk-" + new_key_hash)
+
+    await session.commit()
+
+    response_data = {
+        "api_keys": new_keys,
+        "count": count,
+        "cost_msats": total_cost,
+        "cost_sats": total_cost // 1000,
+        "parent_balance": key.balance,
+        "parent_balance_sats": key.balance // 1000,
+    }
+    logger.debug(f"Child key creation response: {response_data}")
+    return response_data
+
+
 @router.api_route(
     "/{path:path}",
     methods=["GET", "POST", "PUT", "DELETE"],

routstr/core/admin.py

@@ -124,7 +124,7 @@ async def partial_apikeys(request: Request) -> str:
 
     rows = "".join(
         [
-            f"<tr><td>{key.hashed_key}</td><td>{key.balance}</td><td>{key.total_spent}</td><td>{key.total_requests}</td><td>{key.refund_address}</td><td>{fmt_time(key.key_expiry_time)}</td></tr>"
+            f"<tr><td>{key.hashed_key}{' <br><small>(Child of ' + key.parent_key_hash[:8] + '...)</small>' if key.parent_key_hash else ''}</td><td>{key.balance}</td><td>{key.total_spent}</td><td>{key.total_requests}</td><td>{key.refund_address}</td><td>{fmt_time(key.key_expiry_time)}</td></tr>"
             for key in api_keys
         ]
     )
@@ -158,6 +158,7 @@ async def get_temporary_balances_api(request: Request) -> list[dict[str, object]
             "total_requests": key.total_requests,
             "refund_address": key.refund_address,
             "key_expiry_time": key.key_expiry_time,
+            "parent_key_hash": key.parent_key_hash,
         }
         for key in api_keys
     ]
@@ -2515,7 +2516,7 @@ async def get_provider_model(provider_id: int, model_id: str) -> dict[str, objec
                 status_code=404, detail="Model not found for this provider"
             )
         return _row_to_model(
-            row, apply_provider_fee=True, provider_fee=provider.provider_fee
+            row, apply_provider_fee=False, provider_fee=provider.provider_fee
         ).dict()  # type: ignore
 
 
@@ -2597,12 +2598,14 @@ async def create_upstream_provider(
     async with create_session() as session:
         result = await session.exec(
             select(UpstreamProviderRow).where(
-                UpstreamProviderRow.base_url == payload.base_url
+                UpstreamProviderRow.base_url == payload.base_url,
+                UpstreamProviderRow.api_key == payload.api_key,
             )
         )
         if result.first():
             raise HTTPException(
-                status_code=409, detail="Provider with this base URL already exists"
+                status_code=409,
+                detail="Provider with this base URL and API key already exists",
             )
 
         provider = UpstreamProviderRow(
@@ -2726,7 +2729,10 @@ async def get_provider_models(provider_id: int) -> dict[str, object]:
             raise HTTPException(status_code=404, detail="Provider not found")
 
         db_models = await list_models(
-            session=session, upstream_id=provider_id, include_disabled=True
+            session=session,
+            upstream_id=provider_id,
+            include_disabled=True,
+            apply_fees=False,
         )
 
         upstream_models = []
@@ -2734,10 +2740,7 @@ async def get_provider_models(provider_id: int) -> dict[str, object]:
         if upstream_instance:
             try:
                 raw_models = await upstream_instance.fetch_models()
-                upstream_models = [
-                    upstream_instance._apply_provider_fee_to_model(m)
-                    for m in raw_models
-                ]
+                upstream_models = raw_models
             except Exception as e:
                 logger.error(
                     f"Failed to fetch models from {provider.provider_type}: {e}"

routstr/core/db.py

@@ -5,6 +5,7 @@ from typing import AsyncGenerator
 
 from alembic import command
 from alembic.config import Config
+from sqlalchemy import UniqueConstraint
 from sqlalchemy.ext.asyncio.engine import create_async_engine
 from sqlmodel import Field, Relationship, SQLModel, func, select, update
 from sqlmodel.ext.asyncio.session import AsyncSession
@@ -47,6 +48,9 @@ class ApiKey(SQLModel, table=True):  # type: ignore
         default=None,
         description="Currency of the cashu-token",
     )
+    parent_key_hash: str | None = Field(
+        default=None, foreign_key="api_keys.hashed_key", index=True
+    )
 
     @property
     def total_balance(self) -> int:
@@ -108,11 +112,14 @@ class LightningInvoice(SQLModel, table=True):  # type: ignore
 
 class UpstreamProviderRow(SQLModel, table=True):  # type: ignore
     __tablename__ = "upstream_providers"
+    __table_args__ = (
+        UniqueConstraint("base_url", "api_key", name="uq_upstream_providers_base_url_api_key"),
+    )
     id: int | None = Field(default=None, primary_key=True)
     provider_type: str = Field(
         description="Provider type: custom, openai, anthropic, azure, openrouter, etc."
     )
-    base_url: str = Field(unique=True, description="Base URL of the upstream API")
+    base_url: str = Field(description="Base URL of the upstream API")
     api_key: str = Field(description="API key for the upstream provider")
     api_version: str | None = Field(
         default=None, description="API version for Azure OpenAI"

routstr/core/exceptions.py

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

routstr/core/main.py

@@ -13,10 +13,7 @@ from starlette.exceptions import HTTPException
 from ..balance import balance_router, deprecated_wallet_router
 from ..discovery import providers_cache_refresher, providers_router
 from ..nip91 import announce_provider
-from ..payment.models import (
-    models_router,
-    update_sats_pricing,
-)
+from ..payment.models import models_router, update_sats_pricing
 from ..payment.price import update_prices_periodically
 from ..proxy import initialize_upstreams, proxy_router, refresh_model_maps_periodically
 from ..wallet import periodic_payout
@@ -33,9 +30,9 @@ setup_logging()
 logger = get_logger(__name__)
 
 if os.getenv("VERSION_SUFFIX") is not None:
-    __version__ = f"0.2.2-{os.getenv('VERSION_SUFFIX')}"
+    __version__ = f"0.3.0-{os.getenv('VERSION_SUFFIX')}"
 else:
-    __version__ = "0.2.2"
+    __version__ = "0.3.0"
 
 
 @asynccontextmanager
@@ -66,6 +63,11 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
 
                 await reset_all_reserved_balances(session)
 
+        if not s.admin_password:
+            logger.warning(
+                f"Admin password is not set. Visit {s.http_url or 'http://localhost:8000'}/admin to set the password."
+            )
+
         # Apply app metadata from settings
         try:
             app.title = s.name
@@ -186,6 +188,7 @@ async def info() -> dict:
         "mints": global_settings.cashu_mints,
         "http_url": global_settings.http_url,
         "onion_url": global_settings.onion_url,
+        "child_key_cost_msats": global_settings.child_key_cost,
     }
 
 

routstr/core/settings.py

@@ -6,7 +6,7 @@ import os
 from datetime import datetime, timezone
 from typing import Any
 
-from pydantic.v1 import BaseModel, BaseSettings, Field, validator
+from pydantic.v1 import BaseModel, BaseSettings, Field
 from sqlmodel.ext.asyncio.session import AsyncSession
 
 
@@ -37,13 +37,6 @@ class Settings(BaseSettings):
 
     # Cashu
     cashu_mints: list[str] = Field(default_factory=list, env="CASHU_MINTS")
-
-    @validator("cashu_mints", pre=True, each_item=True)
-    def normalize_mint_url(cls, v: str) -> str:
-        if isinstance(v, str):
-            return v.rstrip("/")
-        return v
-
     receive_ln_address: str = Field(default="", env="RECEIVE_LN_ADDRESS")
     primary_mint: str = Field(default="", env="PRIMARY_MINT_URL")
     primary_mint_unit: str = Field(default="sat", env="PRIMARY_MINT_UNIT")
@@ -59,6 +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")
     # 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(
@@ -69,7 +63,7 @@ class Settings(BaseSettings):
     cors_origins: list[str] = Field(default_factory=lambda: ["*"], env="CORS_ORIGINS")
     tor_proxy_url: str = Field(default="socks5://127.0.0.1:9050", env="TOR_PROXY_URL")
     providers_refresh_interval_seconds: int = Field(
-        default=300, env="PROVIDERS_REFRESH_INTERVAL_SECONDS"
+        default=0, env="PROVIDERS_REFRESH_INTERVAL_SECONDS"
     )
     pricing_refresh_interval_seconds: int = Field(
         default=120, env="PRICING_REFRESH_INTERVAL_SECONDS"

routstr/discovery.py

@@ -6,7 +6,7 @@ from typing import Any
 
 import httpx
 import websockets
-from fastapi import APIRouter
+from fastapi import APIRouter, HTTPException
 
 from .core.logging import get_logger
 from .core.settings import settings
@@ -389,6 +389,9 @@ async def get_providers(
     Return cached providers. If include_json, return provider+health; otherwise provider only.
     Optional filter by pubkey.
     """
+    if settings.providers_refresh_interval_seconds == 0:
+        raise HTTPException(status_code=404, detail="Provider discovery is disabled")
+
     cache = await get_cache()
     if not cache:
         await refresh_providers_cache(pubkey=pubkey)

routstr/payment/cost_calculation.py

@@ -15,6 +15,7 @@ class CostData(BaseModel):
     input_msats: int
     output_msats: int
     total_msats: int
+    total_usd: float = 0.0
 
 
 class MaxCostData(CostData):
@@ -61,6 +62,7 @@ async def calculate_cost(  # todo: can be sync
             input_msats=0,
             output_msats=0,
             total_msats=0,
+            total_usd=0.0,
         )
 
     usage_data = response_data["usage"]
@@ -101,6 +103,7 @@ async def calculate_cost(  # todo: can be sync
                 input_msats=-1,  # Cost field doesn't break down by token type
                 output_msats=-1,
                 total_msats=cost_in_msats,
+                total_usd=usd_cost,
             )
         except Exception as e:
             logger.warning(
@@ -210,6 +213,7 @@ async def calculate_cost(  # todo: can be sync
 
     output_msats = round(output_tokens / 1000 * MSATS_PER_1K_OUTPUT_TOKENS, 3)
     token_based_cost = math.ceil(input_msats + output_msats)
+    total_usd = (token_based_cost / 1000.0) * sats_usd_price()
 
     logger.info(
         "Calculated token-based cost",
@@ -219,6 +223,7 @@ async def calculate_cost(  # todo: can be sync
             "input_cost_msats": input_msats,
             "output_cost_msats": output_msats,
             "total_cost_msats": token_based_cost,
+            "total_usd": total_usd,
             "model": response_data.get("model", "unknown"),
         },
     )
@@ -228,4 +233,5 @@ async def calculate_cost(  # todo: can be sync
         input_msats=int(input_msats),
         output_msats=int(output_msats),
         total_msats=token_based_cost,
+        total_usd=total_usd,
     )

routstr/payment/models.py

@@ -230,6 +230,7 @@ async def list_models(
     session: AsyncSession,
     upstream_id: int,
     include_disabled: bool = False,
+    apply_fees: bool = True,
 ) -> list[Model]:
     from sqlmodel import select
 
@@ -247,12 +248,17 @@ async def list_models(
     return [
         _row_to_model(
             r,
-            apply_provider_fee=True,
+            apply_provider_fee=apply_fees,
             provider_fee=providers_by_id[r.upstream_provider_id].provider_fee
             if r.upstream_provider_id in providers_by_id
             else 1.01,
         )
         for r in rows
+        if include_disabled
+        or (
+            r.upstream_provider_id in providers_by_id
+            and providers_by_id[r.upstream_provider_id].enabled
+        )
     ]
 
 
@@ -265,6 +271,8 @@ async def get_model_by_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)
 

routstr/payment/price.py

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

routstr/proxy.py

@@ -16,7 +16,7 @@ from .core.db import (
     create_session,
     get_session,
 )
-from .core.settings import settings
+from .core.exceptions import UpstreamError
 from .payment.helpers import (
     calculate_discounted_max_cost,
     check_token_balance,
@@ -26,14 +26,15 @@ from .payment.helpers import (
 from .payment.models import Model
 from .upstream import BaseUpstreamProvider
 from .upstream.helpers import init_upstreams
-from .wallet import deserialize_token_from_string
 
 logger = get_logger(__name__)
 proxy_router = APIRouter()
 
 _upstreams: list[BaseUpstreamProvider] = []
 _model_instances: dict[str, Model] = {}  # All aliases -> Model
-_provider_map: dict[str, BaseUpstreamProvider] = {}  # All aliases -> Provider
+_provider_map: dict[
+    str, list[BaseUpstreamProvider]
+] = {}  # All aliases -> List[Provider]
 _unique_models: dict[str, Model] = {}  # Unique model.id -> Model (no duplicates)
 
 
@@ -70,8 +71,8 @@ def get_model_instance(model_id: str) -> Model | None:
     return _model_instances.get(model_id.lower())
 
 
-def get_provider_for_model(model_id: str) -> BaseUpstreamProvider | None:
-    """Get UpstreamProvider for model ID from global cache."""
+def get_provider_for_model(model_id: str) -> list[BaseUpstreamProvider] | None:
+    """Get UpstreamProvider list for model ID from global cache."""
     return _provider_map.get(model_id.lower())
 
 
@@ -98,6 +99,8 @@ async def refresh_model_maps() -> None:
     disabled_model_ids: set[str] = set()
 
     for provider in provider_rows:
+        if not provider.enabled:
+            continue
         for model in provider.models:
             if model.enabled:
                 overrides_by_id[model.id] = (model, provider.provider_fee)
@@ -148,32 +151,14 @@ async def proxy(
     else:
         model_id = request_body_dict.get("model", "unknown")
 
-    if "https://testnut.cashu.space" in settings.cashu_mints:
-        try:
-            token_str = None
-            if x_cashu_header := headers.get("x-cashu"):
-                token_str = x_cashu_header
-            elif auth_header := headers.get("authorization"):
-                parts = auth_header.split(" ")
-                if len(parts) > 1 and not parts[1].startswith("sk-"):
-                    token_str = parts[1]
-
-            if token_str:
-                token_obj = deserialize_token_from_string(token_str)
-                if token_obj.mint == "https://testnut.cashu.space":
-                    model_id = "mock/gpt-420-mock"
-                    request_body_dict["model"] = model_id
-        except Exception:
-            pass
-
     model_obj = get_model_instance(model_id)
     if not model_obj:
         return create_error_response(
             "invalid_model", f"Model '{model_id}' not found", 400, request=request
         )
 
-    upstream = get_provider_for_model(model_id)
-    if not upstream:
+    upstreams = get_provider_for_model(model_id)
+    if not upstreams:
         return create_error_response(
             "invalid_model",
             f"No provider found for model '{model_id}'",
@@ -181,6 +166,10 @@ async def proxy(
             request=request,
         )
 
+    # todo figure out cost calculation since fallback provider is usually not the same price
+    # Use first provider for initial checks/cost calculation
+    # primary_upstream = upstreams[0]
+
     _max_cost_for_model = await get_max_cost_for_model(
         model=model_id, session=session, model_obj=model_obj
     )
@@ -190,14 +179,31 @@ async def proxy(
     check_token_balance(headers, request_body_dict, max_cost_for_model)
 
     if x_cashu := headers.get("x-cashu", None):
-        if is_responses_api:
-            return await upstream.handle_x_cashu_responses(
-                request, x_cashu, path, max_cost_for_model, model_obj
-            )
-        else:
-            return await upstream.handle_x_cashu(
-                request, x_cashu, path, max_cost_for_model, model_obj
-            )
+        last_error = None
+        for i, upstream in enumerate(upstreams):
+            try:
+                if is_responses_api:
+                    return await upstream.handle_x_cashu_responses(
+                        request, x_cashu, path, max_cost_for_model, model_obj
+                    )
+                else:
+                    return await upstream.handle_x_cashu(
+                        request, x_cashu, path, max_cost_for_model, model_obj
+                    )
+            except UpstreamError as e:
+                logger.warning(
+                    f"Upstream {upstream.provider_type} failed (x-cashu): {e}"
+                )
+                if i == len(upstreams) - 1:
+                    last_error = e
+                continue
+
+        return create_error_response(
+            "upstream_error",
+            str(last_error) if last_error else "All upstreams failed",
+            502,
+            request=request,
+        )
 
     elif auth := headers.get("authorization", None):
         key = await get_bearer_token_key(headers, path, session, auth)
@@ -212,56 +218,152 @@ async def proxy(
             )
 
         logger.debug("Processing unauthenticated GET request", extra={"path": path})
-        headers = upstream.prepare_headers(dict(request.headers))
-        return await upstream.forward_get_request(request, path, headers)
+
+        last_error_response = None
+        for i, upstream in enumerate(upstreams):
+            try:
+                headers = upstream.prepare_headers(dict(request.headers))
+                response = await upstream.forward_get_request(request, path, headers)
+
+                if response.status_code in [502, 429] and i < len(upstreams) - 1:
+                    error_message = ""
+                    try:
+                        if hasattr(response, "body"):
+                            body_bytes = response.body
+                            data = json.loads(body_bytes)
+                            if "error" in data:
+                                error_data = data["error"]
+                                if isinstance(error_data, dict):
+                                    error_message = error_data.get("message", "")
+                                elif isinstance(error_data, str):
+                                    error_message = error_data
+                    except Exception:
+                        pass
+
+                    await upstream.on_upstream_error_redirect(
+                        response.status_code, error_message
+                    )
+
+                    logger.warning(
+                        f"Upstream {upstream.provider_type} returned {response.status_code} (GET), trying next provider",
+                        extra={
+                            "status_code": response.status_code,
+                            "upstream": upstream.provider_type,
+                        },
+                    )
+                    continue
+                return response
+            except UpstreamError as e:
+                logger.warning(f"Upstream {upstream.provider_type} failed (GET): {e}")
+                if i == len(upstreams) - 1:
+                    last_error_response = create_error_response(
+                        "upstream_error", str(e), 502, request=request
+                    )
+                continue
+        return last_error_response or create_error_response(
+            "upstream_error", "All upstreams failed", 502, request=request
+        )
 
     if request_body_dict:
         await pay_for_request(key, max_cost_for_model, session)
 
-    headers = upstream.prepare_headers(dict(request.headers))
+    for i, upstream in enumerate(upstreams):
+        headers = upstream.prepare_headers(dict(request.headers))
 
-    if is_responses_api:
-        response = await upstream.forward_responses_request(
-            request,
-            path,
-            headers,
-            request_body,
-            key,
-            max_cost_for_model,
-            session,
-            model_obj,
-        )
-    else:
-        response = await upstream.forward_request(
-            request,
-            path,
-            headers,
-            request_body,
-            key,
-            max_cost_for_model,
-            session,
-            model_obj,
-        )
+        try:
+            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,
+                )
 
-    if response.status_code != 200:
-        await revert_pay_for_request(key, session, max_cost_for_model)
-        logger.warning(
-            "Upstream request failed, revert payment",
-            extra={
-                "status_code": response.status_code,
-                "path": path,
-                "key_hash": key.hashed_key[:8] + "...",
-                "key_balance": key.balance,
-                "max_cost_for_model": max_cost_for_model,
-                "upstream_headers": response.headers
-                if hasattr(response, "headers")
-                else None,
-            },
-        )
-        # Return the mapped error response generated earlier rather than masking with 502
-        return response
+            if response.status_code != 200:
+                # Check if we should retry (502 Upstream Error or 429 Rate Limit)
+                should_retry = response.status_code in [502, 429, 400, 401, 403, 404]
+                if should_retry and i < len(upstreams) - 1:
+                    error_message = ""
+                    try:
+                        if hasattr(response, "body"):
+                            body_bytes = response.body
+                            data = json.loads(body_bytes)
+                            if "error" in data:
+                                error_data = data["error"]
+                                if isinstance(error_data, dict):
+                                    error_message = error_data.get("message", "")
+                                elif isinstance(error_data, str):
+                                    error_message = error_data
+                    except Exception:
+                        pass
 
-    return response
+                    await upstream.on_upstream_error_redirect(
+                        response.status_code, error_message
+                    )
+
+                    logger.warning(
+                        f"Upstream {upstream.provider_type} returned {response.status_code}, trying next provider",
+                        extra={
+                            "status_code": response.status_code,
+                            "upstream": upstream.provider_type,
+                        },
+                    )
+                    continue
+
+                # 4xx error (user error), or other non-retryable error, or last provider failed
+                await revert_pay_for_request(key, session, max_cost_for_model)
+                logger.warning(
+                    "Upstream request failed, revert payment",
+                    extra={
+                        "status_code": response.status_code,
+                        "path": path,
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "key_balance": key.balance,
+                        "max_cost_for_model": max_cost_for_model,
+                        "upstream_headers": response.headers
+                        if hasattr(response, "headers")
+                        else None,
+                    },
+                )
+                return response
+
+            return response
+
+        except UpstreamError as e:
+            logger.warning(
+                f"Upstream {upstream.provider_type} failed: {e}",
+                extra={"retry": i < len(upstreams) - 1},
+            )
+
+            # If this was the last provider
+            if i == len(upstreams) - 1:
+                await revert_pay_for_request(key, session, max_cost_for_model)
+                return create_error_response(
+                    "upstream_error", str(e), 502, request=request
+                )
+
+            # Otherwise loop continues to next provider
+            continue
+
+    # Should not be reached given logic above
+    return create_error_response(
+        "upstream_error", "All upstreams failed", 502, request=request
+    )
 
 
 async def get_bearer_token_key(

routstr/upstream/base.py

@@ -15,6 +15,7 @@ from pydantic import BaseModel
 from ..auth import adjust_payment_for_tokens
 from ..core import get_logger
 from ..core.db import ApiKey, AsyncSession, create_session
+from ..core.exceptions import UpstreamError
 
 if TYPE_CHECKING:
     from ..core.db import UpstreamProviderRow
@@ -340,6 +341,20 @@ class BaseUpstreamProvider:
                 message = preview[:500]
         return message, upstream_code
 
+    async def on_upstream_error_redirect(
+        self, status_code: int, error_message: str
+    ) -> None:
+        """Hook called when the proxy redirects to another provider due to an error.
+
+        Subclasses can implement this to perform actions like disabling the provider
+        if it's out of balance.
+
+        Args:
+            status_code: The HTTP status code returned by the upstream
+            error_message: The error message extracted from the upstream response
+        """
+        pass
+
     async def map_upstream_error_response(
         self, request: Request, path: str, upstream_response: httpx.Response
     ) -> Response:
@@ -436,164 +451,111 @@ class BaseUpstreamProvider:
         async def stream_with_cost(
             max_cost_for_model: int,
         ) -> AsyncGenerator[bytes, None]:
-            stored_chunks: list[bytes] = []
             usage_finalized: bool = False
             last_model_seen: str | None = None
+            usage_chunk_data: dict | None = None
+            done_seen: bool = False
 
-            async def finalize_without_usage() -> bytes | None:
+            async def finalize_db_only() -> None:
                 nonlocal usage_finalized
                 if usage_finalized:
-                    return None
+                    return
                 async with create_session() as new_session:
                     fresh_key = await new_session.get(key.__class__, key.hashed_key)
                     if not fresh_key:
-                        logger.warning(
-                            "Key not found when finalizing streaming payment",
-                            extra={"key_hash": key.hashed_key[:8] + "..."},
-                        )
-                        usage_finalized = True
-                        return None
+                        return
                     try:
-                        fallback: dict = {
-                            "model": last_model_seen or "unknown",
-                            "usage": None,
-                        }
-                        cost_data = await adjust_payment_for_tokens(
-                            fresh_key, fallback, new_session, max_cost_for_model
+                        await adjust_payment_for_tokens(
+                            fresh_key,
+                            {"model": last_model_seen or "unknown", "usage": None},
+                            new_session,
+                            max_cost_for_model,
                         )
                         usage_finalized = True
-                        logger.info(
-                            "Finalized streaming payment without explicit usage",
-                            extra={
-                                "key_hash": key.hashed_key[:8] + "...",
-                                "cost_data": cost_data,
-                                "balance_after_adjustment": fresh_key.balance,
-                            },
-                        )
-                        return f"data: {json.dumps({'cost': cost_data})}\n\n".encode()
-                    except Exception as cost_error:
-                        logger.error(
-                            "Error finalizing payment without usage",
-                            extra={
-                                "error": str(cost_error),
-                                "error_type": type(cost_error).__name__,
-                                "key_hash": key.hashed_key[:8] + "...",
-                            },
-                        )
-                        usage_finalized = True
-                        return None
-
-            try:
-                async for chunk in response.aiter_bytes():
-                    stored_chunks.append(chunk)
-                    try:
-                        for part in re.split(b"data: ", chunk):
-                            if not part or part.strip() in (b"[DONE]", b""):
-                                continue
-                            try:
-                                obj = json.loads(part)
-                                if isinstance(obj, dict) and obj.get("model"):
-                                    last_model_seen = str(obj.get("model"))
-                            except json.JSONDecodeError:
-                                pass
                     except Exception:
                         pass
 
-                    yield chunk
+            try:
+                async for chunk in response.aiter_bytes():
+                    # Split chunk into SSE events
+                    parts = re.split(b"data: ", chunk)
+                    for i, part in enumerate(parts):
+                        if not part:
+                            continue
 
-                logger.debug(
-                    "Streaming completed, analyzing usage data",
-                    extra={
-                        "key_hash": key.hashed_key[:8] + "...",
-                        "chunks_count": len(stored_chunks),
-                    },
-                )
+                        stripped_part = part.strip()
+                        if not stripped_part:
+                            continue
 
-                for i in range(len(stored_chunks) - 1, -1, -1):
-                    chunk = stored_chunks[i]
-                    if not chunk:
-                        continue
-                    try:
-                        events = re.split(b"data: ", chunk)
-                        for event_data in events:
-                            if not event_data or event_data.strip() in (b"[DONE]", b""):
-                                continue
-                            try:
-                                data = json.loads(event_data)
-                                if isinstance(data, dict) and data.get("model"):
-                                    last_model_seen = str(data.get("model"))
-                                if isinstance(data, dict) and isinstance(
-                                    data.get("usage"), dict
-                                ):
-                                    async with create_session() as new_session:
-                                        fresh_key = await new_session.get(
-                                            key.__class__, key.hashed_key
-                                        )
-                                        if fresh_key:
-                                            try:
-                                                cost_data = (
-                                                    await adjust_payment_for_tokens(
-                                                        fresh_key,
-                                                        data,
-                                                        new_session,
-                                                        max_cost_for_model,
-                                                    )
-                                                )
-                                                usage_finalized = True
-                                                logger.info(
-                                                    "Payment adjustment completed for streaming",
-                                                    extra={
-                                                        "key_hash": key.hashed_key[:8]
-                                                        + "...",
-                                                        "cost_data": cost_data,
-                                                        "model": last_model_seen,
-                                                        "balance_after_adjustment": fresh_key.balance,
-                                                    },
-                                                )
-                                                yield f"data: {json.dumps({'cost': cost_data})}\n\n".encode()
-                                            except Exception as cost_error:
-                                                logger.error(
-                                                    "Error adjusting payment for streaming tokens",
-                                                    extra={
-                                                        "error": str(cost_error),
-                                                        "error_type": type(
-                                                            cost_error
-                                                        ).__name__,
-                                                        "key_hash": key.hashed_key[:8]
-                                                        + "...",
-                                                    },
-                                                )
-                                    break
-                            except json.JSONDecodeError:
-                                continue
-                    except Exception as e:
-                        logger.error(
-                            "Error processing streaming response chunk",
-                            extra={
-                                "error": str(e),
-                                "error_type": type(e).__name__,
-                                "key_hash": key.hashed_key[:8] + "...",
-                            },
+                        if stripped_part == b"[DONE]":
+                            done_seen = True
+                            continue
+
+                        try:
+                            obj = json.loads(part)
+                            if isinstance(obj, dict):
+                                if obj.get("model"):
+                                    last_model_seen = str(obj.get("model"))
+
+                                if isinstance(obj.get("usage"), dict):
+                                    # Hold this chunk back to merge cost later
+                                    usage_chunk_data = obj
+                                    continue
+                        except json.JSONDecodeError:
+                            pass
+
+                        prefix = (
+                            b"data: " if (i > 0 or chunk.startswith(b"data: ")) else b""
                         )
+                        yield prefix + part
+
+                # Stream finished, process usage if found
+                if usage_chunk_data:
+                    async with create_session() as session:
+                        fresh_key = await session.get(key.__class__, key.hashed_key)
+                        if fresh_key:
+                            try:
+                                cost_data = await adjust_payment_for_tokens(
+                                    fresh_key,
+                                    usage_chunk_data,
+                                    session,
+                                    max_cost_for_model,
+                                )
+                                # Merge cost into usage
+                                usage_chunk_data["usage"]["cost"] = cost_data.get(
+                                    "total_usd", 0.0
+                                )
+                                # Keep detailed cost in metadata
+                                usage_chunk_data["metadata"] = usage_chunk_data.get(
+                                    "metadata", {}
+                                )
+                                usage_chunk_data["metadata"]["routstr"] = {
+                                    "cost": cost_data
+                                }
+                                yield f"data: {json.dumps(usage_chunk_data)}\n\n".encode()
+                                usage_finalized = True
+                            except Exception:
+                                # Fallback: yield original usage chunk if adjustment fails
+                                yield f"data: {json.dumps(usage_chunk_data)}\n\n".encode()
 
                 if not usage_finalized:
-                    maybe_cost_event = await finalize_without_usage()
-                    if maybe_cost_event is not None:
-                        yield maybe_cost_event
+                    await finalize_db_only()
+
+                if done_seen:
+                    yield b"data: [DONE]\n\n"
 
             except Exception as stream_error:
                 logger.warning(
-                    "Streaming interrupted; finalizing without usage",
+                    "Streaming interrupted; finalizing in background",
                     extra={
                         "error": str(stream_error),
-                        "error_type": type(stream_error).__name__,
                         "key_hash": key.hashed_key[:8] + "...",
                     },
                 )
                 raise
             finally:
                 if not usage_finalized:
-                    await finalize_without_usage()
+                    await finalize_db_only()
 
         # Remove inaccurate encoding headers from upstream response
         response_headers = dict(response.headers)
@@ -633,6 +595,7 @@ class BaseUpstreamProvider:
             },
         )
 
+        content: bytes | None = None
         try:
             content = await response.aread()
             response_json = json.loads(content)
@@ -649,6 +612,14 @@ class BaseUpstreamProvider:
             cost_data = await adjust_payment_for_tokens(
                 key, response_json, session, deducted_max_cost
             )
+
+            # Merge cost into usage for OpenCode
+            if "usage" in response_json:
+                response_json["usage"]["cost"] = cost_data.get("total_usd", 0.0)
+
+            # Keep detailed cost
+            response_json["metadata"] = response_json.get("metadata", {})
+            response_json["metadata"]["routstr"] = {"cost": cost_data}
             response_json["cost"] = cost_data
 
             logger.info(
@@ -734,180 +705,135 @@ class BaseUpstreamProvider:
         async def stream_with_responses_cost(
             max_cost_for_model: int,
         ) -> AsyncGenerator[bytes, None]:
-            stored_chunks: list[bytes] = []
             usage_finalized: bool = False
             last_model_seen: str | None = None
             reasoning_tokens: int = 0
+            usage_chunk_data: dict | None = None
+            done_seen: bool = False
 
-            async def finalize_without_usage() -> bytes | None:
+            async def finalize_db_only() -> None:
                 nonlocal usage_finalized
                 if usage_finalized:
-                    return None
+                    return
                 async with create_session() as new_session:
                     fresh_key = await new_session.get(key.__class__, key.hashed_key)
                     if not fresh_key:
-                        logger.warning(
-                            "Key not found when finalizing Responses API streaming payment",
-                            extra={"key_hash": key.hashed_key[:8] + "..."},
-                        )
-                        usage_finalized = True
-                        return None
+                        return
                     try:
-                        fallback: dict = {
-                            "model": last_model_seen or "unknown",
-                            "usage": None,
-                        }
-                        cost_data = await adjust_payment_for_tokens(
-                            fresh_key, fallback, new_session, max_cost_for_model
+                        await adjust_payment_for_tokens(
+                            fresh_key,
+                            {"model": last_model_seen or "unknown", "usage": None},
+                            new_session,
+                            max_cost_for_model,
                         )
                         usage_finalized = True
-                        logger.info(
-                            "Finalized Responses API streaming payment without explicit usage",
-                            extra={
-                                "key_hash": key.hashed_key[:8] + "...",
-                                "cost_data": cost_data,
-                                "balance_after_adjustment": fresh_key.balance,
-                            },
-                        )
-                        return f"data: {json.dumps({'cost': cost_data})}\\n\\n".encode()
-                    except Exception as cost_error:
-                        logger.error(
-                            "Error finalizing Responses API payment without usage",
-                            extra={
-                                "error": str(cost_error),
-                                "error_type": type(cost_error).__name__,
-                                "key_hash": key.hashed_key[:8] + "...",
-                            },
-                        )
-                        usage_finalized = True
-                        return None
-
-            try:
-                async for chunk in response.aiter_bytes():
-                    stored_chunks.append(chunk)
-                    try:
-                        for part in re.split(b"data: ", chunk):
-                            if not part or part.strip() in (b"[DONE]", b""):
-                                continue
-                            try:
-                                obj = json.loads(part)
-                                if isinstance(obj, dict):
-                                    if obj.get("model"):
-                                        last_model_seen = str(obj.get("model"))
-
-                                    # Track reasoning tokens for Responses API
-                                    if usage := obj.get("usage", {}):
-                                        if (
-                                            isinstance(usage, dict)
-                                            and "reasoning_tokens" in usage
-                                        ):
-                                            reasoning_tokens += usage.get(
-                                                "reasoning_tokens", 0
-                                            )
-                            except json.JSONDecodeError:
-                                pass
                     except Exception:
                         pass
 
-                    yield chunk
+            try:
+                async for chunk in response.aiter_bytes():
+                    # Split chunk into SSE events
+                    parts = re.split(b"data: ", chunk)
+                    for i, part in enumerate(parts):
+                        if not part:
+                            continue
 
-                logger.debug(
-                    "Responses API streaming completed, analyzing usage data",
-                    extra={
-                        "key_hash": key.hashed_key[:8] + "...",
-                        "chunks_count": len(stored_chunks),
-                        "reasoning_tokens": reasoning_tokens,
-                    },
-                )
+                        stripped_part = part.strip()
+                        if not stripped_part:
+                            continue
 
-                # Process final usage data
-                for i in range(len(stored_chunks) - 1, -1, -1):
-                    chunk = stored_chunks[i]
-                    if not chunk:
-                        continue
-                    try:
-                        events = re.split(b"data: ", chunk)
-                        for event_data in events:
-                            if not event_data or event_data.strip() in (b"[DONE]", b""):
-                                continue
-                            try:
-                                data = json.loads(event_data)
-                                if isinstance(data, dict) and data.get("model"):
-                                    last_model_seen = str(data.get("model"))
-                                if isinstance(data, dict) and isinstance(
-                                    data.get("usage"), dict
-                                ):
-                                    # Include reasoning tokens in usage calculation
-                                    async with create_session() as new_session:
-                                        fresh_key = await new_session.get(
-                                            key.__class__, key.hashed_key
+                        if stripped_part == b"[DONE]":
+                            done_seen = True
+                            continue
+
+                        try:
+                            obj = json.loads(part)
+                            if isinstance(obj, dict):
+                                if obj.get("model"):
+                                    last_model_seen = str(obj.get("model"))
+
+                                # Track reasoning tokens for Responses API
+                                if usage := obj.get("usage", {}):
+                                    if (
+                                        isinstance(usage, dict)
+                                        and "reasoning_tokens" in usage
+                                    ):
+                                        reasoning_tokens += usage.get(
+                                            "reasoning_tokens", 0
                                         )
-                                        if fresh_key:
-                                            try:
-                                                cost_data = (
-                                                    await adjust_payment_for_tokens(
-                                                        fresh_key,
-                                                        data,
-                                                        new_session,
-                                                        max_cost_for_model,
-                                                    )
-                                                )
-                                                usage_finalized = True
-                                                logger.info(
-                                                    "Payment adjustment completed for Responses API streaming",
-                                                    extra={
-                                                        "key_hash": key.hashed_key[:8]
-                                                        + "...",
-                                                        "cost_data": cost_data,
-                                                        "model": last_model_seen,
-                                                        "reasoning_tokens": reasoning_tokens,
-                                                        "balance_after_adjustment": fresh_key.balance,
-                                                    },
-                                                )
-                                                yield f"data: {json.dumps({'cost': cost_data})}\\n\\n".encode()
-                                            except Exception as cost_error:
-                                                logger.error(
-                                                    "Error adjusting payment for Responses API streaming tokens",
-                                                    extra={
-                                                        "error": str(cost_error),
-                                                        "error_type": type(
-                                                            cost_error
-                                                        ).__name__,
-                                                        "key_hash": key.hashed_key[:8]
-                                                        + "...",
-                                                    },
-                                                )
-                                    break
-                            except json.JSONDecodeError:
-                                continue
-                    except Exception as e:
-                        logger.error(
-                            "Error processing Responses API streaming response chunk",
-                            extra={
-                                "error": str(e),
-                                "error_type": type(e).__name__,
-                                "key_hash": key.hashed_key[:8] + "...",
-                            },
+
+                                # Responses API usage is in response.completed/incomplete events
+                                chunk_type = obj.get("type", "")
+                                if chunk_type in (
+                                    "response.completed",
+                                    "response.incomplete",
+                                ):
+                                    usage_chunk_data = obj
+                                    continue
+                        except json.JSONDecodeError:
+                            pass
+
+                        prefix = (
+                            b"data: " if (i > 0 or chunk.startswith(b"data: ")) else b""
                         )
+                        yield prefix + part
+
+                # Stream finished, process usage if found
+                if usage_chunk_data:
+                    async with create_session() as session:
+                        fresh_key = await session.get(key.__class__, key.hashed_key)
+                        if fresh_key:
+                            try:
+                                cost_data = await adjust_payment_for_tokens(
+                                    fresh_key,
+                                    usage_chunk_data,
+                                    session,
+                                    max_cost_for_model,
+                                )
+                                # Merge cost into usage chunk
+                                if (
+                                    "response" in usage_chunk_data
+                                    and "usage" in usage_chunk_data["response"]
+                                ):
+                                    usage_chunk_data["response"]["usage"]["cost"] = (
+                                        cost_data.get("total_usd", 0.0)
+                                    )
+                                elif "usage" in usage_chunk_data:
+                                    usage_chunk_data["usage"]["cost"] = cost_data.get(
+                                        "total_usd", 0.0
+                                    )
+
+                                # Keep detailed cost in metadata
+                                usage_chunk_data["metadata"] = usage_chunk_data.get(
+                                    "metadata", {}
+                                )
+                                usage_chunk_data["metadata"]["routstr"] = {
+                                    "cost": cost_data
+                                }
+                                yield f"data: {json.dumps(usage_chunk_data)}\n\n".encode()
+                                usage_finalized = True
+                            except Exception:
+                                # Fallback: yield original usage chunk if adjustment fails
+                                yield f"data: {json.dumps(usage_chunk_data)}\n\n".encode()
 
                 if not usage_finalized:
-                    maybe_cost_event = await finalize_without_usage()
-                    if maybe_cost_event is not None:
-                        yield maybe_cost_event
+                    await finalize_db_only()
+
+                if done_seen:
+                    yield b"data: [DONE]\n\n"
 
             except Exception as stream_error:
                 logger.warning(
-                    "Responses API streaming interrupted; finalizing without usage",
+                    "Responses API streaming interrupted; finalizing in background",
                     extra={
                         "error": str(stream_error),
-                        "error_type": type(stream_error).__name__,
                         "key_hash": key.hashed_key[:8] + "...",
                     },
                 )
                 raise
             finally:
                 if not usage_finalized:
-                    await finalize_without_usage()
+                    await finalize_db_only()
 
         # Remove inaccurate encoding headers from upstream response
         response_headers = dict(response.headers)
@@ -947,6 +873,7 @@ class BaseUpstreamProvider:
             },
         )
 
+        content: bytes | None = None
         try:
             content = await response.aread()
             response_json = json.loads(content)
@@ -966,6 +893,14 @@ class BaseUpstreamProvider:
             cost_data = await adjust_payment_for_tokens(
                 key, response_json, session, deducted_max_cost
             )
+
+            # Merge cost into usage for OpenCode
+            if "usage" in response_json:
+                response_json["usage"]["cost"] = cost_data.get("total_usd", 0.0)
+
+            # Keep detailed cost
+            response_json["metadata"] = response_json.get("metadata", {})
+            response_json["metadata"]["routstr"] = {"cost": cost_data}
             response_json["cost"] = cost_data
 
             logger.info(
@@ -1148,6 +1083,14 @@ class BaseUpstreamProvider:
             )
 
             if response.status_code != 200:
+                if response.status_code >= 500:
+                    await response.aclose()
+                    await client.aclose()
+                    raise UpstreamError(
+                        f"Upstream returned status {response.status_code}",
+                        status_code=response.status_code,
+                    )
+
                 try:
                     mapped_error = await self.map_upstream_error_response(
                         request, path, response
@@ -1238,6 +1181,9 @@ class BaseUpstreamProvider:
                 background=background_tasks,
             )
 
+        except UpstreamError:
+            raise
+
         except httpx.RequestError as exc:
             await client.aclose()
             error_type = type(exc).__name__
@@ -1265,9 +1211,7 @@ class BaseUpstreamProvider:
             else:
                 error_message = f"Error connecting to upstream service: {error_type}"
 
-            return create_error_response(
-                "upstream_error", error_message, 502, request=request
-            )
+            raise UpstreamError(error_message, status_code=502)
 
         except Exception as exc:
             await client.aclose()
@@ -1380,6 +1324,14 @@ class BaseUpstreamProvider:
             )
 
             if response.status_code != 200:
+                if response.status_code >= 500:
+                    await response.aclose()
+                    await client.aclose()
+                    raise UpstreamError(
+                        f"Upstream returned status {response.status_code}",
+                        status_code=response.status_code,
+                    )
+
                 try:
                     mapped_error = await self.map_upstream_error_response(
                         request, path, response
@@ -1447,6 +1399,9 @@ class BaseUpstreamProvider:
                 background=background_tasks,
             )
 
+        except UpstreamError:
+            raise
+
         except httpx.RequestError as exc:
             await client.aclose()
             error_type = type(exc).__name__
@@ -1474,9 +1429,7 @@ class BaseUpstreamProvider:
             else:
                 error_message = f"Error connecting to upstream service: {error_type}"
 
-            return create_error_response(
-                "upstream_error", error_message, 502, request=request
-            )
+            raise UpstreamError(error_message, status_code=502)
 
         except Exception as exc:
             await client.aclose()

routstr/upstream/fake.py

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

routstr/upstream/helpers.py

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

routstr/upstream/ppqai.py

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

routstr/wallet.py

@@ -313,8 +313,6 @@ async def periodic_payout() -> None:
         try:
             async with db.create_session() as session:
                 for mint_url in settings.cashu_mints:
-                    if mint_url == "https://testnut.cashu.space":
-                        continue
                     for unit in ["sat", "msat"]:
                         wallet = await get_wallet(mint_url, unit)
                         proofs = get_proofs_per_mint_and_unit(

tests/integration/test_child_keys.py

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

tests/integration/test_example.py

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

tests/integration/test_performance_load.py

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

tests/integration/test_provider_management.py

@@ -3,7 +3,7 @@ Integration tests for provider management functionality.
 Tests GET /v1/providers/ endpoint for listing and managing providers.
 """
 
-from typing import Any
+from typing import Any, Generator
 from unittest.mock import patch
 
 import pytest
@@ -11,7 +11,7 @@ from httpx import AsyncClient
 
 from routstr.discovery import _PROVIDERS_CACHE
 
-from .utils import PerformanceValidator, ResponseValidator
+from .utils import ResponseValidator
 
 
 @pytest.fixture(autouse=True)
@@ -19,6 +19,15 @@ def _clear_providers_cache() -> None:
     _PROVIDERS_CACHE.clear()
 
 
+@pytest.fixture(autouse=True)
+def _enable_provider_discovery() -> Generator[None, Any, Any]:
+    """Enable provider discovery for all tests in this module"""
+    with patch(
+        "routstr.core.settings.settings.providers_refresh_interval_seconds", 300
+    ):
+        yield
+
+
 @pytest.mark.integration
 @pytest.mark.asyncio
 async def test_providers_endpoint_default_response(
@@ -518,46 +527,6 @@ async def test_providers_endpoint_response_format(
             assert isinstance(data_json["providers"], list)
 
 
-@pytest.mark.integration
-@pytest.mark.asyncio
-async def test_providers_endpoint_performance(integration_client: AsyncClient) -> None:
-    """Test providers endpoint meets performance requirements"""
-
-    # Mock quick responses to avoid network delays
-    mock_events: list[dict[str, Any]] = [
-        {
-            "id": f"event{i}",
-            "content": f"Provider: http://provider{i}.onion",
-            "created_at": 1234567890 + i,
-        }
-        for i in range(5)
-    ]
-
-    validator = PerformanceValidator()
-
-    with patch(
-        "routstr.discovery.query_nostr_relay_for_providers", return_value=mock_events
-    ):
-        with patch("routstr.discovery.fetch_provider_health") as mock_fetch:
-            mock_fetch.return_value = {"status_code": 200, "json": {"status": "online"}}
-
-            # Test multiple requests
-            for i in range(10):
-                start = validator.start_timing("providers_endpoint")
-                response = await integration_client.get("/v1/providers/")
-                validator.end_timing("providers_endpoint", start)
-
-                assert response.status_code == 200
-
-    # Validate performance (should be fast with mocked dependencies)
-    perf_result = validator.validate_response_time(
-        "providers_endpoint",
-        max_duration=2.0,  # Allow more time since it involves multiple operations
-        percentile=0.95,
-    )
-    assert perf_result["valid"], f"Performance requirement failed: {perf_result}"
-
-
 @pytest.mark.integration
 @pytest.mark.asyncio
 async def test_providers_endpoint_concurrent_requests(

tests/integration/test_proxy_get_endpoints.py

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

tests/integration/test_proxy_post_endpoints.py

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

tests/integration/test_wallet_information.py

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

tests/integration/test_wallet_refund.py

@@ -537,42 +537,4 @@ async def test_refund_with_expired_key(
         assert response.json()["recipient"] == "expired@ln.address"
 
 
-@pytest.mark.integration
-@pytest.mark.asyncio
-@pytest.mark.slow
-async def test_refund_performance(
-    integration_client: AsyncClient, testmint_wallet: Any
-) -> None:
-    """Test refund endpoint performance"""
 
-    import time
-
-    # Create multiple API keys
-    api_keys = []
-    for i in range(10):
-        token = await testmint_wallet.mint_tokens(100 + i)
-        # Use cashu token as Bearer auth to create API key
-        integration_client.headers["Authorization"] = f"Bearer {token}"
-        response = await integration_client.get("/v1/wallet/info")
-        assert response.status_code == 200
-        api_keys.append(response.json()["api_key"])
-
-    # Measure refund times
-    refund_times = []
-
-    for api_key in api_keys:
-        integration_client.headers["Authorization"] = f"Bearer {api_key}"
-
-        start_time = time.time()
-        response = await integration_client.post("/v1/wallet/refund")
-        end_time = time.time()
-
-        assert response.status_code == 200
-        refund_times.append(end_time - start_time)
-
-    # Performance assertions
-    avg_time = sum(refund_times) / len(refund_times)
-    max_time = max(refund_times)
-
-    assert avg_time < 0.5  # Average under 500ms
-    assert max_time < 1.0  # No refund takes more than 1 second

tests/unit/test_algorithm.py

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

ui/components/AddProviderModelDialog.tsx

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

ui/components/child-key-creator.tsx

@@ -0,0 +1,286 @@
+'use client';
+
+import { useState } from 'react';
+import { WalletService } from '@/lib/api/services/wallet';
+import { Button } from '@/components/ui/button';
+import {
+  Card,
+  CardContent,
+  CardDescription,
+  CardHeader,
+  CardTitle,
+} from '@/components/ui/card';
+import { Alert, AlertDescription, AlertTitle } from '@/components/ui/alert';
+import { Input } from '@/components/ui/input';
+import { Key, Copy, Check, Loader2 } from 'lucide-react';
+import { toast } from 'sonner';
+
+interface ChildKeyCreatorProps {
+  baseUrl?: string;
+  apiKey?: string;
+  onApiKeyChange?: (apiKey: string) => void;
+  costPerKeyMsats?: number;
+}
+
+export function ChildKeyCreator({
+  baseUrl,
+  apiKey: propApiKey,
+  onApiKeyChange,
+  costPerKeyMsats,
+}: ChildKeyCreatorProps) {
+  const [internalApiKey, setInternalApiKey] = useState('');
+  const [loading, setLoading] = useState(false);
+  const [count, setCount] = useState(1);
+  const [newKeys, setNewKeys] = useState<string[]>([]);
+  const [resultInfo, setResultInfo] = useState<{
+    cost_msats: number;
+    parent_balance: number;
+  } | null>(null);
+  const [copiedKey, setCopiedKey] = useState<string | null>(null);
+
+  const activeApiKey = propApiKey ?? internalApiKey;
+
+  const handleApiKeyChange = (val: string) => {
+    setInternalApiKey(val);
+    onApiKeyChange?.(val);
+  };
+
+  const handleCreateKey = async () => {
+    if (!activeApiKey && baseUrl) {
+      toast.error('Please provide a Parent API key first');
+      return;
+    }
+
+    const requestedCount = Math.max(1, Math.min(50, Number(count)));
+
+    setLoading(true);
+    try {
+      const result = await WalletService.createChildKey(
+        baseUrl,
+        activeApiKey,
+        requestedCount
+      );
+
+      console.log('Created child keys:', result);
+
+      if (result.api_keys && result.api_keys.length > 0) {
+        setNewKeys(result.api_keys);
+      } else {
+        throw new Error('No API keys returned from server');
+      }
+
+      setResultInfo({
+        cost_msats: result.cost_msats,
+        parent_balance: result.parent_balance,
+      });
+
+      toast.success(
+        `${requestedCount} child API key${
+          requestedCount > 1 ? 's' : ''
+        } created successfully`
+      );
+    } catch (error) {
+      console.error('Failed to create child key:', error);
+      toast.error(
+        error instanceof Error ? error.message : 'Failed to create child key'
+      );
+    } finally {
+      setLoading(false);
+    }
+  };
+
+  const copyToClipboard = (key: string) => {
+    navigator.clipboard.writeText(key);
+    setCopiedKey(key);
+    toast.success('API key copied to clipboard');
+    setTimeout(() => setCopiedKey(null), 2000);
+  };
+
+  const copyAllToClipboard = () => {
+    navigator.clipboard.writeText(newKeys.join('\n'));
+    toast.success('All API keys copied to clipboard');
+  };
+
+  return (
+    <div className='space-y-6'>
+      <Card>
+        <CardHeader>
+          <div className='flex items-center justify-between'>
+            <div className='space-y-1'>
+              <CardTitle>Create Child API Key</CardTitle>
+              <CardDescription>
+                Generate secondary API keys that share your account balance.
+              </CardDescription>
+            </div>
+            {costPerKeyMsats !== undefined && (
+              <div className='text-right'>
+                <p className='text-muted-foreground text-[0.65rem] tracking-wide uppercase'>
+                  Unit Cost
+                </p>
+                <p className='text-primary text-sm font-bold'>
+                  {costPerKeyMsats / 1000} sats
+                </p>
+              </div>
+            )}
+          </div>
+        </CardHeader>
+        <CardContent>
+          <div className='space-y-4'>
+            {baseUrl && (
+              <div className='space-y-2'>
+                <label className='text-muted-foreground text-[0.7rem] tracking-wider uppercase'>
+                  Parent API Key
+                </label>
+                <Input
+                  value={activeApiKey}
+                  onChange={(e) => handleApiKeyChange(e.target.value)}
+                  placeholder='sk-...'
+                  className='font-mono text-sm'
+                />
+              </div>
+            )}
+
+            <div className='flex flex-col gap-4 sm:flex-row sm:items-end sm:justify-between'>
+              <div className='flex-1 space-y-2'>
+                <div className='flex items-center justify-between'>
+                  <label className='text-muted-foreground text-[0.7rem] tracking-wider uppercase'>
+                    Number of keys
+                  </label>
+                  {costPerKeyMsats && (
+                    <span className='text-muted-foreground text-[10px]'>
+                      Cost: {costPerKeyMsats * count} mSats
+                    </span>
+                  )}
+                </div>
+                <Input
+                  type='number'
+                  min={1}
+                  max={50}
+                  value={count}
+                  onChange={(e) => {
+                    const val = parseInt(e.target.value);
+                    if (!isNaN(val)) {
+                      setCount(Math.max(1, Math.min(50, val)));
+                    } else {
+                      setCount(1);
+                    }
+                  }}
+                  className='w-full sm:w-24'
+                />
+              </div>
+              <Button
+                onClick={handleCreateKey}
+                disabled={loading || (!!baseUrl && !activeApiKey)}
+                className='w-full sm:w-auto'
+              >
+                {loading ? (
+                  <>
+                    <Loader2 className='mr-2 h-4 w-4 animate-spin' />
+                    Creating...
+                  </>
+                ) : (
+                  <>
+                    <Key className='mr-2 h-4 w-4' />
+                    Generate {count > 1 ? `${count} Keys` : 'Key'}
+                  </>
+                )}
+              </Button>
+            </div>
+
+            <p className='text-muted-foreground text-xs'>
+              Each key creation has a small one-time fee.
+            </p>
+
+            {newKeys.length > 0 && (
+              <div className='mt-6 space-y-4'>
+                <Alert className='border-green-200 bg-green-50 dark:border-green-900/20 dark:bg-green-900/10'>
+                  <AlertTitle className='text-green-800 dark:text-green-400'>
+                    {newKeys.length} New API Key{newKeys.length > 1 ? 's' : ''}{' '}
+                    Generated
+                  </AlertTitle>
+                  <AlertDescription className='text-green-700 dark:text-green-500'>
+                    Copy {newKeys.length > 1 ? 'these keys' : 'this key'} now.
+                    You won&apos;t be able to see them again.
+                    {resultInfo && (
+                      <div className='mt-2 font-medium opacity-80'>
+                        Total Cost: {resultInfo.cost_msats / 1000} sats | New
+                        Balance: {resultInfo.parent_balance / 1000} sats
+                      </div>
+                    )}
+                  </AlertDescription>
+                </Alert>
+
+                <div className='space-y-2'>
+                  <div className='flex items-center justify-between'>
+                    <span className='text-muted-foreground text-xs font-medium uppercase'>
+                      Generated Keys ({newKeys.length})
+                    </span>
+                    {newKeys.length > 1 && (
+                      <Button
+                        variant='ghost'
+                        size='sm'
+                        className='h-7 text-[10px] uppercase'
+                        onClick={copyAllToClipboard}
+                      >
+                        <Copy className='mr-1 h-3 w-3' />
+                        Copy All
+                      </Button>
+                    )}
+                  </div>
+                  <div className='grid gap-2'>
+                    {newKeys.map((key, index) => (
+                      <div
+                        key={index}
+                        className='group relative flex items-center gap-2'
+                      >
+                        <code className='bg-muted/50 flex-1 rounded border p-2.5 font-mono text-[10px] break-all sm:text-xs'>
+                          {key}
+                        </code>
+                        <Button
+                          size='icon'
+                          variant='ghost'
+                          className='h-8 w-8 shrink-0'
+                          onClick={() => copyToClipboard(key)}
+                        >
+                          {copiedKey === key ? (
+                            <Check className='h-3.5 w-3.5 text-green-500' />
+                          ) : (
+                            <Copy className='h-3.5 w-3.5 opacity-50 group-hover:opacity-100' />
+                          )}
+                        </Button>
+                      </div>
+                    ))}
+                  </div>
+                </div>
+
+                {newKeys.length > 3 && (
+                  <div className='space-y-2'>
+                    <label className='text-muted-foreground text-[0.7rem] tracking-wider uppercase'>
+                      Bulk Export (All Keys)
+                    </label>
+                    <div className='relative'>
+                      <textarea
+                        readOnly
+                        value={newKeys.join('\n')}
+                        rows={Math.min(newKeys.length, 6)}
+                        className='bg-muted/30 w-full rounded-md border p-3 font-mono text-[10px] focus:outline-none'
+                      />
+                      <Button
+                        size='sm'
+                        variant='secondary'
+                        className='absolute right-2 bottom-2 h-7 text-[10px]'
+                        onClick={copyAllToClipboard}
+                      >
+                        Copy Bulk
+                      </Button>
+                    </div>
+                  </div>
+                )}
+              </div>
+            )}
+          </div>
+        </CardContent>
+      </Card>
+    </div>
+  );
+}

ui/components/landing/cheat-sheet.tsx

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

ui/components/temporary-balances.tsx

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

ui/lib/api/services/admin.ts

@@ -139,18 +139,26 @@ export class AdminService {
   ): Record<string, unknown> {
     if (!pricing) return pricing;
     const result = { ...pricing };
-    if (typeof result.prompt === 'number') {
-      result.prompt = result.prompt * 1000000;
-    }
-    if (typeof result.completion === 'number') {
-      result.completion = result.completion * 1000000;
-    }
-    if (typeof result.request === 'number') {
-      result.request = result.request * 1000000;
-    }
-    if (typeof result.image === 'number') {
-      result.image = result.image * 1000000;
-    }
+
+    // Only prompt and completion are per-token and need scaling to per-1M
+    const convertField = (field: string) => {
+      const val = result[field];
+      if (val !== undefined && val !== null) {
+        const num = typeof val === 'string' ? parseFloat(val) : (val as number);
+        if (!isNaN(num)) {
+          // Multiply by 1M and round to avoid floating point artifacts (e.g. 0.40399999999999997)
+          // 9 decimals is plenty for USD/1M tokens (0.000000001)
+          result[field] = parseFloat((num * 1000000).toFixed(9));
+        }
+      }
+    };
+
+    convertField('prompt');
+    convertField('completion');
+
+    // Other fields (request, image, etc.) are already flat fees (per item)
+    // so we do NOT scale them.
+
     return result;
   }
 
@@ -159,18 +167,23 @@ export class AdminService {
   ): Record<string, unknown> {
     if (!pricing) return pricing;
     const result = { ...pricing };
-    if (typeof result.prompt === 'number') {
-      result.prompt = result.prompt / 1000000;
-    }
-    if (typeof result.completion === 'number') {
-      result.completion = result.completion / 1000000;
-    }
-    if (typeof result.request === 'number') {
-      result.request = result.request / 1000000;
-    }
-    if (typeof result.image === 'number') {
-      result.image = result.image / 1000000;
-    }
+
+    // Only prompt and completion are per-1M in UI and need scaling down to per-token
+    const convertField = (field: string) => {
+      const val = result[field];
+      if (val !== undefined && val !== null) {
+        const num = typeof val === 'string' ? parseFloat(val) : (val as number);
+        if (!isNaN(num)) {
+          result[field] = num / 1000000;
+        }
+      }
+    };
+
+    convertField('prompt');
+    convertField('completion');
+
+    // Other fields stay as flat fees
+
     return result;
   }
 
@@ -291,7 +304,19 @@ export class AdminService {
     const data = await apiClient.get<ProviderModels>(
       `/admin/api/upstream-providers/${providerId}/models`
     );
-    return data;
+
+    // Convert pricing for all models in the list so the UI receives "per 1M tokens" values
+    return {
+      ...data,
+      db_models: data.db_models.map((m) => ({
+        ...m,
+        pricing: this.convertPricingToPerMillionTokens(m.pricing),
+      })),
+      remote_models: data.remote_models.map((m) => ({
+        ...m,
+        pricing: this.convertPricingToPerMillionTokens(m.pricing),
+      })),
+    };
   }
 
   static async createProviderModel(
@@ -498,8 +523,8 @@ export class AdminService {
       : null;
 
     const pricing = {
-      prompt: (data.input_cost as number) / 1000000,
-      completion: (data.output_cost as number) / 1000000,
+      prompt: data.input_cost as number,
+      completion: data.output_cost as number,
       request: (data.min_cost_per_request as number) || 0,
       image: 0,
       web_search: 0,
@@ -553,8 +578,8 @@ export class AdminService {
     const existingModel = await this.getModel(modelId, providerId);
 
     const pricing = {
-      prompt: (data.input_cost as number) / 1000000,
-      completion: (data.output_cost as number) / 1000000,
+      prompt: data.input_cost as number,
+      completion: data.output_cost as number,
       request: (data.min_cost_per_request as number) || 0,
       image: 0,
       web_search: 0,
@@ -926,6 +951,7 @@ export const TemporaryBalanceSchema = z.object({
   total_requests: z.number(),
   refund_address: z.string().nullable(),
   key_expiry_time: z.number().nullable(),
+  parent_key_hash: z.string().nullable().optional(),
 });
 
 export type TemporaryBalance = z.infer<typeof TemporaryBalanceSchema>;

ui/lib/api/services/wallet.ts

@@ -29,6 +29,28 @@ export type RedeemTokenResponse = z.infer<typeof RedeemTokenResponseSchema>;
 export type SendTokenRequest = z.infer<typeof SendTokenRequestSchema>;
 export type SendTokenResponse = z.infer<typeof SendTokenResponseSchema>;
 
+export interface BalanceDetail {
+  mint_url: string;
+  unit: string;
+  wallet_balance: number;
+  user_balance: number;
+  owner_balance: number;
+  error?: string;
+}
+
+export interface WithdrawResponse {
+  token: string;
+}
+
+export interface CreateChildKeyResponse {
+  api_keys: string[];
+  count: number;
+  cost_msats: number;
+  cost_sats: number;
+  parent_balance: number;
+  parent_balance_sats: number;
+}
+
 export class WalletService {
   static async redeemToken(token: string): Promise<RedeemTokenResponse> {
     try {
@@ -108,17 +130,38 @@ export class WalletService {
       throw error;
     }
   }
-}
 
-export interface BalanceDetail {
-  mint_url: string;
-  unit: string;
-  wallet_balance: number;
-  user_balance: number;
-  owner_balance: number;
-  error?: string;
-}
+  static async createChildKey(
+    baseUrl?: string,
+    apiKey?: string,
+    count: number = 1
+  ): Promise<CreateChildKeyResponse> {
+    try {
+      if (baseUrl && apiKey) {
+        const response = await fetch(`${baseUrl}/v1/balance/child-key`, {
+          method: 'POST',
+          headers: {
+            'Content-Type': 'application/json',
+            Authorization: `Bearer ${apiKey}`,
+          },
+          body: JSON.stringify({ count }),
+        });
 
-export interface WithdrawResponse {
-  token: string;
+        if (!response.ok) {
+          const errorText = await response.text();
+          throw new Error(errorText || 'Failed to create child key');
+        }
+
+        return (await response.json()) as CreateChildKeyResponse;
+      }
+
+      return await apiClient.post<CreateChildKeyResponse>(
+        '/v1/balance/child-key',
+        { count }
+      );
+    } catch (error) {
+      console.error('Error creating child key:', error);
+      throw error;
+    }
+  }
 }

uv.lock

@@ -1878,7 +1878,7 @@ wheels = [
 
 [[package]]
 name = "routstr"
-version = "0.2.2"
+version = "0.3.0"
 source = { editable = "." }
 dependencies = [
     { name = "aiosqlite" },