remdb 0.3.180__py3-none-any.whl → 0.3.230__py3-none-any.whl

rem/api/routers/admin.py

@@ -31,6 +31,8 @@ from fastapi import APIRouter, Depends, Header, HTTPException, Query, Background
 from loguru import logger
 from pydantic import BaseModel
 
+from .common import ErrorResponse
+
 from ..deps import require_admin
 from ...models.entities import Message, Session, SessionMode
 from ...services.postgres import Repository
@@ -103,7 +105,13 @@ class SystemStats(BaseModel):
 # =============================================================================
 
 
-@router.get("/users", response_model=UserListResponse)
+@router.get(
+    "/users",
+    response_model=UserListResponse,
+    responses={
+        503: {"model": ErrorResponse, "description": "Database not enabled"},
+    },
+)
 async def list_all_users(
     user: dict = Depends(require_admin),
     limit: int = Query(default=50, ge=1, le=100),
@@ -155,7 +163,13 @@ async def list_all_users(
     return UserListResponse(data=summaries, total=total, has_more=has_more)
 
 
-@router.get("/sessions", response_model=SessionListResponse)
+@router.get(
+    "/sessions",
+    response_model=SessionListResponse,
+    responses={
+        503: {"model": ErrorResponse, "description": "Database not enabled"},
+    },
+)
 async def list_all_sessions(
     user: dict = Depends(require_admin),
     user_id: str | None = Query(default=None, description="Filter by user ID"),
@@ -202,7 +216,13 @@ async def list_all_sessions(
     return SessionListResponse(data=sessions, total=total, has_more=has_more)
 
 
-@router.get("/messages", response_model=MessageListResponse)
+@router.get(
+    "/messages",
+    response_model=MessageListResponse,
+    responses={
+        503: {"model": ErrorResponse, "description": "Database not enabled"},
+    },
+)
 async def list_all_messages(
     user: dict = Depends(require_admin),
     user_id: str | None = Query(default=None, description="Filter by user ID"),
@@ -252,7 +272,13 @@ async def list_all_messages(
     return MessageListResponse(data=messages, total=total, has_more=has_more)
 
 
-@router.get("/stats", response_model=SystemStats)
+@router.get(
+    "/stats",
+    response_model=SystemStats,
+    responses={
+        503: {"model": ErrorResponse, "description": "Database not enabled"},
+    },
+)
 async def get_system_stats(
     user: dict = Depends(require_admin),
 ) -> SystemStats:

rem/api/routers/auth.py

@@ -101,6 +101,8 @@ from authlib.integrations.starlette_client import OAuth
 from pydantic import BaseModel, EmailStr
 from loguru import logger
 
+from .common import ErrorResponse
+
 from ...settings import settings
 from ...services.postgres.service import PostgresService
 from ...services.user_service import UserService
@@ -159,7 +161,14 @@ class EmailVerifyRequest(BaseModel):
     code: str
 
 
-@router.post("/email/send-code")
+@router.post(
+    "/email/send-code",
+    responses={
+        400: {"model": ErrorResponse, "description": "Invalid request or email rejected"},
+        500: {"model": ErrorResponse, "description": "Failed to send login code"},
+        501: {"model": ErrorResponse, "description": "Email auth or database not configured"},
+    },
+)
 async def send_email_code(request: Request, body: EmailSendCodeRequest):
     """
     Send a login code to an email address.
@@ -221,7 +230,14 @@ async def send_email_code(request: Request, body: EmailSendCodeRequest):
         await db.disconnect()
 
 
-@router.post("/email/verify")
+@router.post(
+    "/email/verify",
+    responses={
+        400: {"model": ErrorResponse, "description": "Invalid or expired code"},
+        500: {"model": ErrorResponse, "description": "Failed to verify login code"},
+        501: {"model": ErrorResponse, "description": "Email auth or database not configured"},
+    },
+)
 async def verify_email_code(request: Request, body: EmailVerifyRequest):
     """
     Verify login code and create session with JWT tokens.
@@ -319,7 +335,13 @@ async def verify_email_code(request: Request, body: EmailVerifyRequest):
 # =============================================================================
 
 
-@router.get("/{provider}/login")
+@router.get(
+    "/{provider}/login",
+    responses={
+        400: {"model": ErrorResponse, "description": "Unknown OAuth provider"},
+        501: {"model": ErrorResponse, "description": "Authentication is disabled"},
+    },
+)
 async def login(provider: str, request: Request):
     """
     Initiate OAuth flow with provider.
@@ -361,7 +383,13 @@ async def login(provider: str, request: Request):
     return await client.authorize_redirect(request, redirect_uri)
 
 
-@router.get("/{provider}/callback")
+@router.get(
+    "/{provider}/callback",
+    responses={
+        400: {"model": ErrorResponse, "description": "Authentication failed or unknown provider"},
+        501: {"model": ErrorResponse, "description": "Authentication is disabled"},
+    },
+)
 async def callback(provider: str, request: Request):
     """
     OAuth callback endpoint.
@@ -498,7 +526,12 @@ async def logout(request: Request):
     return {"message": "Logged out successfully"}
 
 
-@router.get("/me")
+@router.get(
+    "/me",
+    responses={
+        401: {"model": ErrorResponse, "description": "Not authenticated"},
+    },
+)
 async def me(request: Request):
     """
     Get current user information from session or JWT.
@@ -536,11 +569,19 @@ class TokenRefreshRequest(BaseModel):
     refresh_token: str
 
 
-@router.post("/token/refresh")
+@router.post(
+    "/token/refresh",
+    responses={
+        401: {"model": ErrorResponse, "description": "Invalid or expired refresh token"},
+    },
+)
 async def refresh_token(body: TokenRefreshRequest):
     """
     Refresh access token using refresh token.
 
+    Fetches the user's current role/tier from the database to ensure
+    the new access token reflects their actual permissions.
+
     Args:
         body: TokenRefreshRequest with refresh_token
 
@@ -548,7 +589,46 @@ async def refresh_token(body: TokenRefreshRequest):
         New access token or 401 if refresh token is invalid
     """
     jwt_service = get_jwt_service()
-    result = jwt_service.refresh_access_token(body.refresh_token)
+
+    # First decode the refresh token to get user_id (without full verification yet)
+    payload = jwt_service.decode_without_verification(body.refresh_token)
+    if not payload:
+        raise HTTPException(
+            status_code=401,
+            detail="Invalid refresh token format"
+        )
+
+    user_id = payload.get("sub")
+    if not user_id:
+        raise HTTPException(
+            status_code=401,
+            detail="Invalid refresh token: missing user ID"
+        )
+
+    # Fetch user from database to get current role/tier
+    user_override = None
+    if settings.postgres.enabled:
+        db = PostgresService()
+        try:
+            await db.connect()
+            user_service = UserService(db)
+            user_entity = await user_service.get_user_by_id(user_id)
+            if user_entity:
+                user_override = {
+                    "role": user_entity.role or "user",
+                    "roles": [user_entity.role] if user_entity.role else ["user"],
+                    "tier": user_entity.tier.value if user_entity.tier else "free",
+                    "name": user_entity.name,
+                }
+                logger.debug(f"Refresh token: fetched user {user_id} with role={user_override['role']}, tier={user_override['tier']}")
+        except Exception as e:
+            logger.warning(f"Could not fetch user for token refresh: {e}")
+            # Continue without override - will use defaults
+        finally:
+            await db.disconnect()
+
+    # Now do the actual refresh with proper verification
+    result = jwt_service.refresh_access_token(body.refresh_token, user_override=user_override)
 
     if not result:
         raise HTTPException(
@@ -559,7 +639,12 @@ async def refresh_token(body: TokenRefreshRequest):
     return result
 
 
-@router.post("/token/verify")
+@router.post(
+    "/token/verify",
+    responses={
+        401: {"model": ErrorResponse, "description": "Missing, invalid, or expired token"},
+    },
+)
 async def verify_token(request: Request):
     """
     Verify an access token is valid.
@@ -623,7 +708,12 @@ def verify_dev_token(token: str) -> bool:
     return token == expected
 
 
-@router.get("/dev/token")
+@router.get(
+    "/dev/token",
+    responses={
+        401: {"model": ErrorResponse, "description": "Dev tokens not available in production"},
+    },
+)
 async def get_dev_token(request: Request):
     """
     Get a development token for testing (non-production only).
@@ -659,7 +749,13 @@ async def get_dev_token(request: Request):
     }
 
 
-@router.get("/dev/mock-code/{email}")
+@router.get(
+    "/dev/mock-code/{email}",
+    responses={
+        401: {"model": ErrorResponse, "description": "Mock codes not available in production"},
+        404: {"model": ErrorResponse, "description": "No code found for email"},
+    },
+)
 async def get_mock_code(email: str, request: Request):
     """
     Get the mock login code for testing (non-production only).

rem/api/routers/chat/child_streaming.py

@@ -0,0 +1,379 @@
+"""
+Child Agent Event Handling.
+
+Handles events from child agents during multi-agent orchestration.
+
+Event Flow:
+```
+Parent Agent (Siggy)
+      │
+      ▼
+  ask_agent tool
+      │
+      ├──────────────────────────────────┐
+      ▼                                  │
+  Child Agent (intake_diverge)           │
+      │                                  │
+      ├── child_tool_start ──────────────┼──► Event Sink (Queue)
+      ├── child_content ─────────────────┤
+      └── child_tool_result ─────────────┘
+                                         │
+                                         ▼
+                            drain_child_events()
+                                         │
+                                         ├── SSE to client
+                                         └── DB persistence
+```
+
+IMPORTANT: When child_content is streamed, parent text output should be SKIPPED
+to prevent content duplication.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import uuid
+from typing import TYPE_CHECKING, Any, AsyncGenerator
+
+from loguru import logger
+
+from .streaming_utils import StreamingState, build_content_chunk
+from .sse_events import MetadataEvent, ToolCallEvent, format_sse_event
+from ....services.session import SessionMessageStore
+from ....settings import settings
+from ....utils.date_utils import to_iso, utc_now
+
+if TYPE_CHECKING:
+    from ....agentic.context import AgentContext
+
+
+async def handle_child_tool_start(
+    state: StreamingState,
+    child_agent: str,
+    tool_name: str,
+    arguments: dict | None,
+    session_id: str | None,
+    user_id: str | None,
+) -> AsyncGenerator[str, None]:
+    """
+    Handle child_tool_start event.
+
+    Actions:
+    1. Log the tool call
+    2. Emit SSE event
+    3. Save to database
+    """
+    full_tool_name = f"{child_agent}:{tool_name}"
+    tool_id = f"call_{uuid.uuid4().hex[:8]}"
+
+    # Normalize arguments
+    if not isinstance(arguments, dict):
+        arguments = None
+
+    # 1. LOG
+    logger.info(f"🔧 {full_tool_name}")
+
+    # 2. EMIT SSE
+    yield format_sse_event(ToolCallEvent(
+        tool_name=full_tool_name,
+        tool_id=tool_id,
+        status="started",
+        arguments=arguments,
+    ))
+
+    # 3. SAVE TO DB
+    if session_id and settings.postgres.enabled:
+        try:
+            store = SessionMessageStore(
+                user_id=user_id or settings.test.effective_user_id
+            )
+            tool_msg = {
+                "role": "tool",
+                "tool_name": full_tool_name,
+                "content": json.dumps(arguments) if arguments else "",
+                "timestamp": to_iso(utc_now()),
+            }
+            await store.store_session_messages(
+                session_id=session_id,
+                messages=[tool_msg],
+                user_id=user_id,
+                compress=False,
+            )
+        except Exception as e:
+            logger.warning(f"Failed to save child tool call: {e}")
+
+
+def handle_child_content(
+    state: StreamingState,
+    child_agent: str,
+    content: str,
+) -> str | None:
+    """
+    Handle child_content event.
+
+    CRITICAL: Sets state.child_content_streamed = True
+    This flag is used to skip parent text output and prevent duplication.
+
+    Returns:
+        SSE chunk or None if content is empty
+    """
+    if not content:
+        return None
+
+    # Track that child content was streamed
+    # Parent text output should be SKIPPED when this is True
+    state.child_content_streamed = True
+    state.responding_agent = child_agent
+
+    return build_content_chunk(state, content)
+
+
+async def handle_child_tool_result(
+    state: StreamingState,
+    child_agent: str,
+    result: Any,
+    message_id: str | None,
+    session_id: str | None,
+    agent_schema: str | None,
+) -> AsyncGenerator[str, None]:
+    """
+    Handle child_tool_result event.
+
+    Actions:
+    1. Log metadata if present
+    2. Emit metadata event if present
+    3. Emit tool completion event
+    """
+    # Check for metadata registration
+    if isinstance(result, dict) and result.get("_metadata_event"):
+        risk = result.get("risk_level", "")
+        conf = result.get("confidence", "")
+        logger.info(f"📊 {child_agent} metadata: risk={risk}, confidence={conf}")
+
+        # Update responding agent from child
+        if result.get("agent_schema"):
+            state.responding_agent = result.get("agent_schema")
+
+        # Build extra dict with risk fields
+        extra_data = {}
+        if risk:
+            extra_data["risk_level"] = risk
+
+        yield format_sse_event(MetadataEvent(
+            message_id=message_id,
+            session_id=session_id,
+            agent_schema=agent_schema,
+            responding_agent=state.responding_agent,
+            confidence=result.get("confidence"),
+            extra=extra_data if extra_data else None,
+        ))
+
+    # Emit tool completion
+    yield format_sse_event(ToolCallEvent(
+        tool_name=f"{child_agent}:tool",
+        tool_id=f"call_{uuid.uuid4().hex[:8]}",
+        status="completed",
+        result=str(result)[:200] if result else None,
+    ))
+
+
+async def drain_child_events(
+    event_sink: asyncio.Queue,
+    state: StreamingState,
+    session_id: str | None = None,
+    user_id: str | None = None,
+    message_id: str | None = None,
+    agent_schema: str | None = None,
+) -> AsyncGenerator[str, None]:
+    """
+    Drain all pending child events from the event sink.
+
+    This is called during tool execution to process events
+    pushed by child agents via ask_agent.
+
+    IMPORTANT: When child_content events are processed, this sets
+    state.child_content_streamed = True. Callers should check this
+    flag and skip parent text output to prevent duplication.
+    """
+    while not event_sink.empty():
+        try:
+            child_event = event_sink.get_nowait()
+            async for chunk in process_child_event(
+                child_event, state, session_id, user_id, message_id, agent_schema
+            ):
+                yield chunk
+        except Exception as e:
+            logger.warning(f"Error processing child event: {e}")
+
+
+async def process_child_event(
+    child_event: dict,
+    state: StreamingState,
+    session_id: str | None = None,
+    user_id: str | None = None,
+    message_id: str | None = None,
+    agent_schema: str | None = None,
+) -> AsyncGenerator[str, None]:
+    """Process a single child event and yield SSE chunks."""
+    event_type = child_event.get("type", "")
+    child_agent = child_event.get("agent_name", "child")
+
+    if event_type == "child_tool_start":
+        async for chunk in handle_child_tool_start(
+            state=state,
+            child_agent=child_agent,
+            tool_name=child_event.get("tool_name", "tool"),
+            arguments=child_event.get("arguments"),
+            session_id=session_id,
+            user_id=user_id,
+        ):
+            yield chunk
+
+    elif event_type == "child_content":
+        chunk = handle_child_content(
+            state=state,
+            child_agent=child_agent,
+            content=child_event.get("content", ""),
+        )
+        if chunk:
+            yield chunk
+
+    elif event_type == "child_tool_result":
+        async for chunk in handle_child_tool_result(
+            state=state,
+            child_agent=child_agent,
+            result=child_event.get("result"),
+            message_id=message_id,
+            session_id=session_id,
+            agent_schema=agent_schema,
+        ):
+            yield chunk
+
+
+async def stream_with_child_events(
+    tools_stream,
+    child_event_sink: asyncio.Queue,
+    state: StreamingState,
+    session_id: str | None = None,
+    user_id: str | None = None,
+    message_id: str | None = None,
+    agent_schema: str | None = None,
+) -> AsyncGenerator[tuple[str, Any], None]:
+    """
+    Multiplex tool events with child events using asyncio.wait().
+
+    This is the key fix for child agent streaming - instead of draining
+    the queue synchronously during tool event iteration, we concurrently
+    listen to both sources and yield events as they arrive.
+
+    Yields:
+        Tuples of (event_type, event_data) where event_type is either
+        "tool" or "child", allowing the caller to handle each appropriately.
+    """
+    tool_iter = tools_stream.__aiter__()
+
+    # Create initial tasks
+    pending_tool: asyncio.Task | None = None
+    pending_child: asyncio.Task | None = None
+
+    try:
+        pending_tool = asyncio.create_task(tool_iter.__anext__())
+    except StopAsyncIteration:
+        # No tool events, just drain any remaining child events
+        while not child_event_sink.empty():
+            try:
+                child_event = child_event_sink.get_nowait()
+                yield ("child", child_event)
+            except asyncio.QueueEmpty:
+                break
+        return
+
+    # Start listening for child events with a short timeout
+    pending_child = asyncio.create_task(
+        _get_child_event_with_timeout(child_event_sink, timeout=0.05)
+    )
+
+    try:
+        while True:
+            # Wait for either source to produce an event
+            tasks = {t for t in [pending_tool, pending_child] if t is not None}
+            if not tasks:
+                break
+
+            done, _ = await asyncio.wait(tasks, return_when=asyncio.FIRST_COMPLETED)
+
+            for task in done:
+                try:
+                    result = task.result()
+                except asyncio.TimeoutError:
+                    # Child queue timeout - restart listener
+                    if task is pending_child:
+                        pending_child = asyncio.create_task(
+                            _get_child_event_with_timeout(child_event_sink, timeout=0.05)
+                        )
+                    continue
+                except StopAsyncIteration:
+                    # Tool stream exhausted
+                    if task is pending_tool:
+                        pending_tool = None
+                        # Final drain of any remaining child events
+                        if pending_child:
+                            pending_child.cancel()
+                            try:
+                                await pending_child
+                            except asyncio.CancelledError:
+                                pass
+                        while not child_event_sink.empty():
+                            try:
+                                child_event = child_event_sink.get_nowait()
+                                yield ("child", child_event)
+                            except asyncio.QueueEmpty:
+                                break
+                        return
+                    continue
+
+                if task is pending_child and result is not None:
+                    # Got a child event
+                    yield ("child", result)
+                    # Restart child listener
+                    pending_child = asyncio.create_task(
+                        _get_child_event_with_timeout(child_event_sink, timeout=0.05)
+                    )
+                elif task is pending_tool:
+                    # Got a tool event
+                    yield ("tool", result)
+                    # Get next tool event
+                    try:
+                        pending_tool = asyncio.create_task(tool_iter.__anext__())
+                    except StopAsyncIteration:
+                        pending_tool = None
+                elif task is pending_child and result is None:
+                    # Timeout with no event - restart listener
+                    pending_child = asyncio.create_task(
+                        _get_child_event_with_timeout(child_event_sink, timeout=0.05)
+                    )
+    finally:
+        # Cleanup any pending tasks
+        for task in [pending_tool, pending_child]:
+            if task and not task.done():
+                task.cancel()
+                try:
+                    await task
+                except asyncio.CancelledError:
+                    pass
+
+
+async def _get_child_event_with_timeout(
+    queue: asyncio.Queue, timeout: float = 0.05
+) -> dict | None:
+    """
+    Get an event from the queue with a timeout.
+
+    Returns None on timeout (no event available).
+    This allows the multiplexer to check for tool events regularly.
+    """
+    try:
+        return await asyncio.wait_for(queue.get(), timeout=timeout)
+    except asyncio.TimeoutError:
+        return None