Merge pull request #53 from basedosdados/staging

chore: sync main with staging

Author: vrtornisiello

.gitignore

@@ -1,4 +1,5 @@
-# Claude
+# Agents
+.agents/
 .claude/
 
 # Python-generated files

alembic/versions/19e2c92563e2_update_message_status.py

@@ -0,0 +1,41 @@
+"""Add MODEL_CALL_LIMIT and INTERRUPTED message statuses.
+
+Revision ID: 19e2c92563e2
+Revises: 1c6556bb74f2
+Create Date: 2026-05-14 16:02:45.217643
+"""
+
+from typing import Sequence, Union
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "19e2c92563e2"
+down_revision: Union[str, Sequence[str], None] = "1c6556bb74f2"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    """Add MODEL_CALL_LIMIT and INTERRUPTED to the messagestatus enum.
+
+    Note: `ALTER TYPE ... ADD VALUE` cannot run inside a transaction block
+    on PostgreSQL pre-v12, so we use an autocommit block.
+    """
+    with op.get_context().autocommit_block():
+        op.execute(
+            "ALTER TYPE messagestatus ADD VALUE IF NOT EXISTS 'MODEL_CALL_LIMIT'"
+        )
+        op.execute("ALTER TYPE messagestatus ADD VALUE IF NOT EXISTS 'INTERRUPTED'")
+
+
+def downgrade() -> None:
+    """Downgrade is intentionally unsupported.
+
+    Postgres cannot drop enum values, and rebuilding the type would require
+    remapping existing MODEL_CALL_LIMIT and INTERRUPTED rows to another status,
+    losing the diagnostic signal these statuses were added to capture.
+    """
+    raise NotImplementedError(
+        "Downgrade not supported: removing enum values would silently rewrite rows."
+    )

app/api/dependencies/__init__.py

@@ -1,4 +1,4 @@
-from .agent import Agent, get_agent
+from .agent import Agent, RunningRuns, get_agent, get_running_runs
 from .auth import UserID, get_user_id
 from .db import AsyncDB, get_database, get_session
 from .feedback import FeedbackSender, get_feedback_sender
@@ -7,10 +7,12 @@
     "Agent",
     "AsyncDB",
     "FeedbackSender",
+    "RunningRuns",
     "UserID",
     "get_agent",
     "get_database",
     "get_feedback_sender",
+    "get_running_runs",
     "get_session",
     "get_user_id",
 ]

app/api/dependencies/agent.py

@@ -1,3 +1,4 @@
+import asyncio
 from typing import Annotated
 
 from fastapi import Depends, Request
@@ -8,4 +9,9 @@ def get_agent(request: Request) -> CompiledStateGraph:
     return request.app.state.agent
 
 
+def get_running_runs(request: Request) -> dict[str, asyncio.Task]:
+    return request.app.state.running_runs
+
+
 Agent = Annotated[CompiledStateGraph, Depends(get_agent)]
+RunningRuns = Annotated[dict[str, asyncio.Task], Depends(get_running_runs)]

app/api/routers/chatbot.py

@@ -3,10 +3,12 @@
 
 from fastapi import APIRouter, BackgroundTasks, HTTPException, status
 from fastapi.responses import StreamingResponse
+from loguru import logger
 
-from app.api.dependencies import Agent, AsyncDB, FeedbackSender, UserID
+from app.api.dependencies import Agent, AsyncDB, FeedbackSender, RunningRuns, UserID
 from app.api.schemas import ConfigDict, UserMessage
-from app.api.streaming import stream_response
+from app.api.streaming import run_agent, stream_events
+from app.api.streaming.schemas import StreamEvent
 from app.db.models import (
     FeedbackCreate,
     FeedbackPayload,
@@ -96,14 +98,19 @@ async def list_messages(
     return await database.get_messages(thread.id, order_by)
 
 
-@router.post("/threads/{thread_id}/messages")
+@router.post(
+    "/threads/{thread_id}/messages",
+    response_class=StreamingResponse,
+    status_code=status.HTTP_201_CREATED,
+)
 async def send_message(
     thread_id: str,
     user_message: UserMessage,
-    agent: Agent,
     database: AsyncDB,
+    agent: Agent,
+    running_runs: RunningRuns,
     user_id: UserID,
-) -> Message:
+) -> StreamingResponse:
     run_id = str(uuid.uuid4())
 
     config = ConfigDict(
@@ -120,15 +127,35 @@ async def send_message(
 
     message = await database.create_message(message_create)
 
-    return StreamingResponse(
-        stream_response(
-            database=database,
+    queue: asyncio.Queue[StreamEvent] = asyncio.Queue()
+
+    task = asyncio.create_task(
+        run_agent(
             agent=agent,
-            user_message=message,
             config=config,
             thread_id=thread_id,
+            user_message=message,
             model_uri=settings.MODEL_URI,
+            queue=queue,
         ),
+        name=f"run_agent:{run_id}",
+    )
+
+    running_runs[run_id] = task
+
+    def _cleanup(task: asyncio.Task):  # pragma: no cover
+        del running_runs[run_id]
+        if task.cancelled():
+            logger.warning(f"run_agent task {run_id} was cancelled mid-run")
+            return
+        e = task.exception()
+        if e is not None:
+            logger.opt(exception=e).error(f"run_agent task {run_id} crashed mid-run:")
+
+    task.add_done_callback(_cleanup)
+
+    return StreamingResponse(
+        stream_events(queue),
         status_code=status.HTTP_201_CREATED,
     )
 

app/api/streaming/__init__.py

@@ -1,3 +1,4 @@
-from app.api.streaming.stream import stream_response
+from app.api.streaming.agent_runner import run_agent
+from app.api.streaming.stream import stream_events
 
-__all__ = ["stream_response"]
+__all__ = ["run_agent", "stream_events"]