Merge branch 'v0.3.0' into claude-code

# Conflicts:
#	routstr/upstream/base.py

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/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/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

@@ -1064,6 +1064,170 @@ class BaseUpstreamProvider:
                     },
                 )
 
+    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,
@@ -1157,7 +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:

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)
 

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;
   }
 

uv.lock

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