Merge branch 'v0.3.0' into claude-code

# Conflicts:
#	routstr/upstream/base.py

.github/workflows/test.yml

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

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "routstr"
-version = "0.2.1"
+version = "0.2.2"
 description = "Payment proxy for your LLM endpoint using cashu and nostr."
 readme = "README.md"
 requires-python = ">=3.11"

routstr/auth.py

@@ -441,6 +441,29 @@ async def adjust_payment_for_tokens(
         },
     )
 
+    async def release_reservation_only() -> None:
+        """Fallback to release reservation without charging when main update fails."""
+        try:
+            release_stmt = (
+                update(ApiKey)
+                .where(col(ApiKey.hashed_key) == 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()
+            logger.warning(
+                "Released reservation without charging (fallback)",
+                extra={
+                    "key_hash": 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] + "..."},
+            )
+
     match await calculate_cost(response_data, deducted_max_cost, session):
         case MaxCostData() as cost:
             logger.debug(
@@ -465,7 +488,7 @@ async def adjust_payment_for_tokens(
             await session.commit()
             if result.rowcount == 0:
                 logger.error(
-                    "Failed to finalize max-cost payment - insufficient reserved balance",
+                    "Failed to finalize max-cost payment - retrying reservation release",
                     extra={
                         "key_hash": key.hashed_key[:8] + "...",
                         "deducted_max_cost": deducted_max_cost,
@@ -474,6 +497,7 @@ async def adjust_payment_for_tokens(
                         "model": model,
                     },
                 )
+                await release_reservation_only()
             else:
                 await session.refresh(key)
                 logger.info(
@@ -568,13 +592,14 @@ async def adjust_payment_for_tokens(
                     )
                 else:
                     logger.warning(
-                        "Failed to finalize additional charge (concurrent operation)",
+                        "Failed to finalize additional charge - releasing reservation",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
                             "attempted_charge": total_cost_msats,
                             "model": model,
                         },
                     )
+                    await release_reservation_only()
             else:
                 # Refund some of the base cost
                 refund = abs(cost_difference)
@@ -603,7 +628,7 @@ async def adjust_payment_for_tokens(
 
                 if result.rowcount == 0:
                     logger.error(
-                        "Failed to finalize payment - insufficient reserved balance",
+                        "Failed to finalize payment - releasing reservation",
                         extra={
                             "key_hash": key.hashed_key[:8] + "...",
                             "deducted_max_cost": deducted_max_cost,
@@ -612,28 +637,27 @@ async def adjust_payment_for_tokens(
                             "model": model,
                         },
                     )
-                    # Still return the cost data even if we couldn't properly finalize
-                    # The reservation was already made, so the user has paid
+                    await release_reservation_only()
+                else:
+                    cost.total_msats = total_cost_msats
+                    await session.refresh(key)
 
-                cost.total_msats = total_cost_msats
-                await session.refresh(key)
-
-                logger.info(
-                    "Refund processed successfully",
-                    extra={
-                        "key_hash": key.hashed_key[:8] + "...",
-                        "refunded_amount": refund,
-                        "new_balance": key.balance,
-                        "final_cost": cost.total_msats,
-                        "model": model,
-                    },
-                )
+                    logger.info(
+                        "Refund processed successfully",
+                        extra={
+                            "key_hash": key.hashed_key[:8] + "...",
+                            "refunded_amount": refund,
+                            "new_balance": key.balance,
+                            "final_cost": cost.total_msats,
+                            "model": model,
+                        },
+                    )
 
             return cost.dict()
 
         case CostDataError() as error:
             logger.error(
-                "Cost calculation error during payment adjustment",
+                "Cost calculation error during payment adjustment - releasing reservation",
                 extra={
                     "key_hash": key.hashed_key[:8] + "...",
                     "model": model,
@@ -641,6 +665,7 @@ async def adjust_payment_for_tokens(
                     "error_code": error.code,
                 },
             )
+            await release_reservation_only()
 
             raise HTTPException(
                 status_code=400,
@@ -652,7 +677,12 @@ async def adjust_payment_for_tokens(
                     }
                 },
             )
-    # Fallback return to satisfy type checker; execution should not reach here
+    # Fallback: should not reach here, but release reservation just in case
+    logger.error(
+        "Unexpected fallback in adjust_payment_for_tokens - releasing reservation",
+        extra={"key_hash": key.hashed_key[:8] + "...", "model": model},
+    )
+    await release_reservation_only()
     return {
         "base_msats": deducted_max_cost,
         "input_msats": 0,

routstr/balance.py

@@ -154,7 +154,8 @@ async def refund_wallet_endpoint(
         return cached
 
     key: ApiKey = await validate_bearer_key(bearer_value, session)
-    remaining_balance_msats: int = key.balance
+
+    remaining_balance_msats: int = key.total_balance
 
     if key.refund_currency == "sat":
         remaining_balance = remaining_balance_msats // 1000

routstr/core/db.py

@@ -6,7 +6,7 @@ from typing import AsyncGenerator
 from alembic import command
 from alembic.config import Config
 from sqlalchemy.ext.asyncio.engine import create_async_engine
-from sqlmodel import Field, Relationship, SQLModel, func, select
+from sqlmodel import Field, Relationship, SQLModel, func, select, update
 from sqlmodel.ext.asyncio.session import AsyncSession
 
 from .logging import get_logger
@@ -53,6 +53,14 @@ class ApiKey(SQLModel, table=True):  # type: ignore
         return self.balance - self.reserved_balance
 
 
+async def reset_all_reserved_balances(session: AsyncSession) -> None:
+    logger.info("Resetting all reserved balances to 0")
+    stmt = update(ApiKey).values(reserved_balance=0)
+    await session.exec(stmt)  # type: ignore[call-overload]
+    await session.commit()
+    logger.info("Reserved balances reset successfully")
+
+
 class ModelRow(SQLModel, table=True):  # type: ignore
     __tablename__ = "models"
     id: str = Field(primary_key=True)

routstr/core/main.py

@@ -33,9 +33,9 @@ setup_logging()
 logger = get_logger(__name__)
 
 if os.getenv("VERSION_SUFFIX") is not None:
-    __version__ = f"0.2.1-{os.getenv('VERSION_SUFFIX')}"
+    __version__ = f"0.2.2-{os.getenv('VERSION_SUFFIX')}"
 else:
-    __version__ = "0.2.1"
+    __version__ = "0.2.2"
 
 
 @asynccontextmanager
@@ -61,6 +61,10 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
         # Initialize application settings (env -> computed -> DB precedence)
         async with create_session() as session:
             s = await SettingsService.initialize(session)
+            if s.reset_reserved_balance_on_startup:
+                from .db import reset_all_reserved_balances
+
+                await reset_all_reserved_balances(session)
 
         # Apply app metadata from settings
         try:

routstr/core/settings.py

@@ -61,6 +61,9 @@ class Settings(BaseSettings):
     tolerance_percentage: float = Field(default=1.0, env="TOLERANCE_PERCENTAGE")
     # Minimum per-request charge in millisatoshis when model pricing is free/zero
     min_request_msat: int = Field(default=1, env="MIN_REQUEST_MSAT")
+    reset_reserved_balance_on_startup: bool = Field(
+        default=True, env="RESET_RESERVED_BALANCE_ON_STARTUP"
+    )  # deactivate in horizontal scaling setups
 
     # Network
     cors_origins: list[str] = Field(default_factory=lambda: ["*"], env="CORS_ORIGINS")

routstr/payment/cost_calculation.py

@@ -48,13 +48,6 @@ async def calculate_cost(  # todo: can be sync
         },
     )
 
-    cost_data = MaxCostData(
-        base_msats=max_cost,
-        input_msats=0,
-        output_msats=0,
-        total_msats=max_cost,
-    )
-
     if "usage" not in response_data or response_data["usage"] is None:
         logger.warning(
             "No usage data in response, using base cost only",
@@ -63,7 +56,12 @@ async def calculate_cost(  # todo: can be sync
                 "model": response_data.get("model", "unknown"),
             },
         )
-        return cost_data
+        return MaxCostData(
+            base_msats=0,
+            input_msats=0,
+            output_msats=0,
+            total_msats=0,
+        )
 
     usage_data = response_data["usage"]
 
@@ -178,7 +176,12 @@ async def calculate_cost(  # todo: can be sync
                 "model": response_data.get("model", "unknown"),
             },
         )
-        return cost_data
+        return MaxCostData(
+            base_msats=max_cost,
+            input_msats=0,
+            output_msats=0,
+            total_msats=max_cost,
+        )
 
     input_tokens = usage_data.get("prompt_tokens", 0)
     output_tokens = usage_data.get("completion_tokens", 0)

routstr/payment/lnurl.py

@@ -283,7 +283,7 @@ async def raw_send_to_lnurl(
             f"({min_sendable_sat} - {max_sendable_sat} {unit})"
         )
 
-    estimated_fees_sat = int(max(math.ceil((amount_msat / 1000) * 0.01), 2))
+    estimated_fees_sat = int(max(math.ceil((amount_msat / 1000) * 0.01), 2)) + 1
     estimated_fees_msat = estimated_fees_sat * 1000
     final_amount = amount_msat - estimated_fees_msat
 

routstr/upstream/base.py

@@ -447,6 +447,11 @@ class BaseUpstreamProvider:
                 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
                     try:
                         fallback: dict = {
@@ -475,6 +480,7 @@ class BaseUpstreamProvider:
                                 "key_hash": key.hashed_key[:8] + "...",
                             },
                         )
+                        usage_finalized = True
                         return None
 
             try:
@@ -584,8 +590,10 @@ class BaseUpstreamProvider:
                         "key_hash": key.hashed_key[:8] + "...",
                     },
                 )
-                await finalize_without_usage()
                 raise
+            finally:
+                if not usage_finalized:
+                    await finalize_without_usage()
 
         # Remove inaccurate encoding headers from upstream response
         response_headers = dict(response.headers)
@@ -738,6 +746,11 @@ class BaseUpstreamProvider:
                 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
                     try:
                         fallback: dict = {
@@ -766,6 +779,7 @@ class BaseUpstreamProvider:
                                 "key_hash": key.hashed_key[:8] + "...",
                             },
                         )
+                        usage_finalized = True
                         return None
 
             try:
@@ -890,8 +904,10 @@ class BaseUpstreamProvider:
                         "key_hash": key.hashed_key[:8] + "...",
                     },
                 )
-                await finalize_without_usage()
                 raise
+            finally:
+                if not usage_finalized:
+                    await finalize_without_usage()
 
         # Remove inaccurate encoding headers from upstream response
         response_headers = dict(response.headers)
@@ -1010,6 +1026,208 @@ class BaseUpstreamProvider:
             )
             raise
 
+    async def _finalize_generic_streaming_payment(
+        self, key_hash: str, max_cost: int, path: str
+    ) -> None:
+        """Background task to finalize payment for generic streaming requests."""
+        async with create_session() as session:
+            key = await session.get(ApiKey, key_hash)
+            if not key:
+                logger.warning(
+                    "Key not found during background payment finalization",
+                    extra={"key_hash": key_hash[:8] + "..."},
+                )
+                return
+
+            try:
+                # Finalize with "unknown" model and no usage to release reservation/charge max cost
+                await adjust_payment_for_tokens(
+                    key,
+                    {"model": "unknown", "usage": None},
+                    session,
+                    max_cost,
+                )
+                logger.info(
+                    "Finalized generic streaming payment in background",
+                    extra={
+                        "path": path,
+                        "key_hash": key_hash[:8] + "...",
+                    },
+                )
+            except Exception as e:
+                logger.error(
+                    "Error finalizing generic streaming payment in background",
+                    extra={
+                        "error": str(e),
+                        "key_hash": key_hash[:8] + "...",
+                        "path": path,
+                    },
+                )
+
+    async def handle_streaming_messages_completion(
+        self, response: httpx.Response, key: ApiKey, max_cost_for_model: int
+    ) -> StreamingResponse:
+        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
+            input_tokens: int = 0
+            output_tokens: int = 0
+
+            async def finalize_without_usage() -> bytes | None:
+                nonlocal usage_finalized
+                if usage_finalized:
+                    return None
+                async with create_session() as new_session:
+                    fresh_key = await new_session.get(key.__class__, key.hashed_key)
+                    if not fresh_key:
+                        usage_finalized = True
+                        return None
+                    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
+                        )
+                        usage_finalized = True
+                        return f"event: cost\ndata: {json.dumps({'cost': cost_data})}\n\n".encode()
+                    except Exception:
+                        usage_finalized = True
+                        return None
+
+            try:
+                async for chunk in response.aiter_bytes():
+                    stored_chunks.append(chunk)
+                    try:
+                        decoded_chunk = chunk.decode("utf-8", errors="ignore")
+                        for line in decoded_chunk.split("\n"):
+                            if line.startswith("data: "):
+                                try:
+                                    data = json.loads(line[6:])
+                                    if isinstance(data, dict):
+                                        msg = data.get("message", {})
+                                        if msg and msg.get("model"):
+                                            last_model_seen = str(msg.get("model"))
+
+                                        if usage := msg.get("usage"):
+                                            input_tokens += usage.get("input_tokens", 0)
+                                            output_tokens += usage.get(
+                                                "output_tokens", 0
+                                            )
+
+                                        if usage := data.get("usage"):
+                                            input_tokens += usage.get("input_tokens", 0)
+                                            output_tokens += usage.get(
+                                                "output_tokens", 0
+                                            )
+                                except json.JSONDecodeError:
+                                    pass
+                    except Exception:
+                        pass
+
+                    yield chunk
+
+                usage_data = {
+                    "input_tokens": input_tokens,
+                    "output_tokens": output_tokens,
+                }
+
+                if input_tokens > 0 or output_tokens > 0:
+                    async with create_session() as new_session:
+                        fresh_key = await new_session.get(key.__class__, key.hashed_key)
+                        if fresh_key:
+                            try:
+                                combined_data = {
+                                    "model": last_model_seen or "unknown",
+                                    "usage": usage_data,
+                                }
+                                cost_data = await adjust_payment_for_tokens(
+                                    fresh_key,
+                                    combined_data,
+                                    new_session,
+                                    max_cost_for_model,
+                                )
+                                usage_finalized = True
+                                yield f"event: cost\ndata: {json.dumps({'cost': cost_data})}\n\n".encode()
+                            except Exception:
+                                pass
+
+                if not usage_finalized:
+                    maybe_cost_event = await finalize_without_usage()
+                    if maybe_cost_event is not None:
+                        yield maybe_cost_event
+
+            except Exception:
+                if not usage_finalized:
+                    await finalize_without_usage()
+                raise
+            finally:
+                if not usage_finalized:
+                    await finalize_without_usage()
+
+        response_headers = dict(response.headers)
+        response_headers.pop("content-encoding", None)
+        response_headers.pop("content-length", None)
+
+        return StreamingResponse(
+            stream_with_cost(max_cost_for_model),
+            status_code=response.status_code,
+            headers=response_headers,
+        )
+
+    async def handle_non_streaming_messages_completion(
+        self,
+        response: httpx.Response,
+        key: ApiKey,
+        session: AsyncSession,
+        deducted_max_cost: int,
+        path: str,
+    ) -> Response:
+        try:
+            content = await response.aread()
+            response_json = json.loads(content)
+
+            if path.endswith("count_tokens") and "usage" not in response_json:
+                input_tokens = response_json.get("input_tokens", 0)
+                response_json["usage"] = {"input_tokens": input_tokens}
+
+            cost_data = await adjust_payment_for_tokens(
+                key, response_json, session, deducted_max_cost
+            )
+            response_json["cost"] = cost_data
+
+            allowed_headers = {
+                "content-type",
+                "cache-control",
+                "date",
+                "vary",
+                "access-control-allow-origin",
+                "access-control-allow-methods",
+                "access-control-allow-headers",
+                "access-control-allow-credentials",
+                "access-control-expose-headers",
+                "access-control-max-age",
+            }
+
+            response_headers = {
+                k: v
+                for k, v in response.headers.items()
+                if k.lower() in allowed_headers
+            }
+
+            return Response(
+                content=json.dumps(response_json).encode(),
+                status_code=response.status_code,
+                headers=response_headers,
+                media_type="application/json",
+            )
+        except Exception:
+            raise
+
     async def forward_request(
         self,
         request: Request,
@@ -1103,7 +1321,54 @@ class BaseUpstreamProvider:
                     await client.aclose()
                 return mapped_error
 
-            if path.endswith("chat/completions") or path.endswith("embeddings"):
+            if (
+                path.endswith("chat/completions")
+                or path.endswith("embeddings")
+                or path.endswith("messages")
+                or path.endswith("messages/count_tokens")
+            ):
+                if path.endswith("messages"):
+                    client_wants_streaming = False
+                    if request_body:
+                        try:
+                            request_data = json.loads(request_body)
+                            client_wants_streaming = request_data.get("stream", False)
+                        except json.JSONDecodeError:
+                            pass
+
+                    content_type = response.headers.get("content-type", "")
+                    upstream_is_streaming = "text/event-stream" in content_type
+                    is_streaming = client_wants_streaming and upstream_is_streaming
+
+                    if is_streaming and response.status_code == 200:
+                        result = await self.handle_streaming_messages_completion(
+                            response, key, max_cost_for_model
+                        )
+                        background_tasks = BackgroundTasks()
+                        background_tasks.add_task(response.aclose)
+                        background_tasks.add_task(client.aclose)
+                        result.background = background_tasks
+                        return result
+
+                    if response.status_code == 200:
+                        try:
+                            return await self.handle_non_streaming_messages_completion(
+                                response, key, session, max_cost_for_model, path
+                            )
+                        finally:
+                            await response.aclose()
+                            await client.aclose()
+
+                if path.endswith("messages/count_tokens"):
+                    if response.status_code == 200:
+                        try:
+                            return await self.handle_non_streaming_messages_completion(
+                                response, key, session, max_cost_for_model, path
+                            )
+                        finally:
+                            await response.aclose()
+                            await client.aclose()
+
                 if path.endswith("chat/completions"):
                     client_wants_streaming = False
                     if request_body:
@@ -1161,6 +1426,12 @@ class BaseUpstreamProvider:
             background_tasks = BackgroundTasks()
             background_tasks.add_task(response.aclose)
             background_tasks.add_task(client.aclose)
+            background_tasks.add_task(
+                self._finalize_generic_streaming_payment,
+                key.hashed_key,
+                max_cost_for_model,
+                path,
+            )
 
             logger.debug(
                 "Streaming non-chat response",
@@ -1364,9 +1635,15 @@ class BaseUpstreamProvider:
             background_tasks = BackgroundTasks()
             background_tasks.add_task(response.aclose)
             background_tasks.add_task(client.aclose)
+            background_tasks.add_task(
+                self._finalize_generic_streaming_payment,
+                key.hashed_key,
+                max_cost_for_model,
+                path,
+            )
 
             logger.debug(
-                "Streaming non-chat response",
+                "Streaming non-Responses API response",
                 extra={
                     "path": path,
                     "status_code": response.status_code,

routstr/upstream/gemini.py

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

routstr/wallet.py

@@ -81,7 +81,7 @@ async def swap_to_primary_mint(
         amount_msat = token_amount
     else:
         raise ValueError("Invalid unit")
-    estimated_fee_sat = math.ceil(max(amount_msat // 1000 * 0.01, 2))
+    estimated_fee_sat = math.ceil(max(amount_msat // 1000 * 0.01, 2)) + 1
     amount_msat_after_fee = amount_msat - estimated_fee_sat * 1000
     primary_wallet = await get_wallet(settings.primary_mint, settings.primary_mint_unit)
 

scripts/reproduce_bug_staging.py

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

ui/Dockerfile

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

ui/Dockerfile.build

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

ui/app/providers/page.tsx

@@ -208,7 +208,12 @@ function ProviderBalance({
     return <Skeleton className='h-9 w-24' />;
   }
 
-  if (error || !balanceData?.ok || !balanceData.balance_data) {
+  if (
+    error ||
+    !balanceData?.ok ||
+    balanceData.balance_data === undefined ||
+    balanceData.balance_data === null
+  ) {
     return null;
   }
 

ui/package.json

@@ -15,6 +15,7 @@
     "@dnd-kit/core": "^6.3.1",
     "@dnd-kit/modifiers": "^9.0.0",
     "@dnd-kit/sortable": "^10.0.0",
+    "@dnd-kit/utilities": "^3.2.2",
     "@hookform/resolvers": "^5.0.1",
     "@radix-ui/react-alert-dialog": "^1.1.10",
     "@radix-ui/react-avatar": "^1.1.6",
@@ -40,17 +41,17 @@
     "@radix-ui/react-toggle": "^1.1.6",
     "@radix-ui/react-toggle-group": "^1.1.6",
     "@radix-ui/react-tooltip": "^1.2.3",
-    "@tanstack/react-query": "^5.74.4",
+    "@tanstack/react-query": "^5.90.16",
     "@tanstack/react-table": "^8.21.3",
     "axios": "^1.13.2",
     "class-variance-authority": "^0.7.1",
     "clsx": "^2.1.1",
     "cmdk": "^1.1.1",
-    "date-fns": "^3.6.0",
+    "date-fns": "^4.1.0",
     "embla-carousel-react": "^8.6.0",
     "input-otp": "^1.4.2",
-    "lucide-react": "^0.501.0",
-    "next": "15.3.1",
+    "lucide-react": "^0.562.0",
+    "next": "15.5.9",
     "next-themes": "^0.4.6",
     "qrcode": "^1.5.4",
     "qrcode.react": "^4.2.0",
@@ -67,21 +68,21 @@
     "zustand": "^5.0.3"
   },
   "devDependencies": {
-    "@eslint/eslintrc": "^3",
-    "@tailwindcss/postcss": "^4",
-    "@tanstack/react-query-devtools": "^5.74.4",
+    "@eslint/eslintrc": "^3.3.3",
+    "@tailwindcss/postcss": "^4.1.18",
+    "@tanstack/react-query-devtools": "^5.91.2",
     "@types/node": "^20",
     "@types/qrcode": "^1.5.6",
     "@types/react": "^19",
     "@types/react-dom": "^19",
     "eslint": "^9.25.0",
-    "eslint-config-next": "15.3.1",
+    "eslint-config-next": "15.5.9",
     "eslint-config-prettier": "^10.1.2",
     "eslint-plugin-prettier": "^5.2.6",
     "eslint-plugin-react": "^7.37.5",
     "prettier": "^3.5.3",
     "prettier-plugin-tailwindcss": "^0.6.11",
-    "tailwindcss": "^4",
+    "tailwindcss": "^4.1.18",
     "typescript": "^5"
   }
 }

ui/tsconfig.json

@@ -22,6 +22,12 @@
       "@/*": ["./*"]
     }
   },
-  "include": ["next-env.d.ts", "**/*.ts", "**/*.tsx", ".next/types/**/*.ts"],
+  "include": [
+    "next-env.d.ts",
+    "**/*.ts",
+    "**/*.tsx",
+    ".next/types/**/*.ts",
+    ".next/dev/types/**/*.ts"
+  ],
   "exclude": ["node_modules"]
 }

uv.lock

@@ -1878,7 +1878,7 @@ wheels = [
 
 [[package]]
 name = "routstr"
-version = "0.2.1"
+version = "0.2.2"
 source = { editable = "." }
 dependencies = [
     { name = "aiosqlite" },