remdb 0.3.181__py3-none-any.whl → 0.3.223__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/completions.py

@@ -215,7 +215,7 @@ async def ensure_session_with_metadata(
     Merges request metadata with existing session metadata.
 
     Args:
-        session_id: Session identifier (maps to Session.name)
+        session_id: Session UUID from X-Session-Id header
         user_id: User identifier
         tenant_id: Tenant identifier
         is_eval: Whether this is an evaluation session
@@ -228,12 +228,8 @@ async def ensure_session_with_metadata(
     try:
         repo = Repository(Session, table_name="sessions")
 
-        # Try to load existing session by name (session_id is the name field)
-        existing_list = await repo.find(
-            filters={"name": session_id, "tenant_id": tenant_id},
-            limit=1,
-        )
-        existing = existing_list[0] if existing_list else None
+        # Look up session by UUID (id field)
+        existing = await repo.get_by_id(session_id)
 
         if existing:
             # Merge metadata if provided
@@ -254,9 +250,10 @@ async def ensure_session_with_metadata(
                 await repo.upsert(existing)
                 logger.debug(f"Updated session {session_id} (eval={is_eval}, metadata keys={list(merged_metadata.keys())})")
         else:
-            # Create new session
+            # Create new session with the provided UUID as the id
             session = Session(
-                name=session_id,
+                id=session_id,  # Use the provided UUID as session id
+                name=session_id,  # Default name to UUID, can be updated later with LLM-generated name
                 mode=SessionMode.EVALUATION if is_eval else SessionMode.NORMAL,
                 user_id=user_id,
                 tenant_id=tenant_id,
@@ -503,16 +500,51 @@ async def chat_completions(body: ChatCompletionRequest, request: Request):
             logger.error(f"Failed to transcribe audio: {e}")
             # Fall through with original content (will likely fail at agent)
 
-    # Use ContextBuilder to construct complete message list with:
-    # 1. System context hint (date + user profile)
-    # 2. Session history (if session_id provided)
-    # 3. New messages from request body (transcribed if audio)
+    # Use ContextBuilder to construct context and basic messages
+    # Note: We load session history separately for proper pydantic-ai message_history
     context, messages = await ContextBuilder.build_from_headers(
         headers=dict(request.headers),
         new_messages=new_messages,
         user_id=temp_context.user_id,  # From JWT token (source of truth)
     )
 
+    # Load raw session history for proper pydantic-ai message_history format
+    # This enables proper tool call/return pairing for LLM API compatibility
+    from ....services.session import SessionMessageStore, session_to_pydantic_messages, audit_session_history
+    from ....agentic.schema import get_system_prompt
+
+    pydantic_message_history = None
+    if context.session_id and settings.postgres.enabled:
+        try:
+            store = SessionMessageStore(user_id=context.user_id or settings.test.effective_user_id)
+            raw_session_history = await store.load_session_messages(
+                session_id=context.session_id,
+                user_id=context.user_id,
+                compress_on_load=False,  # Don't compress - we need full data for reconstruction
+            )
+            if raw_session_history:
+                # CRITICAL: Extract and pass the agent's system prompt
+                # pydantic-ai only auto-adds system prompts when message_history is empty
+                # When we pass message_history, we must include the system prompt ourselves
+                agent_system_prompt = get_system_prompt(agent_schema) if agent_schema else None
+                pydantic_message_history = session_to_pydantic_messages(
+                    raw_session_history,
+                    system_prompt=agent_system_prompt,
+                )
+                logger.debug(f"Converted {len(raw_session_history)} session messages to {len(pydantic_message_history)} pydantic-ai messages (with system prompt)")
+
+                # Audit session history if enabled (for debugging)
+                audit_session_history(
+                    session_id=context.session_id,
+                    agent_name=schema_name or "default",
+                    prompt=body.messages[-1].content if body.messages else "",
+                    raw_session_history=raw_session_history,
+                    pydantic_messages_count=len(pydantic_message_history),
+                )
+        except Exception as e:
+            logger.warning(f"Failed to load session history for message_history: {e}")
+            # Fall back to old behavior (concatenated prompt)
+
     logger.info(f"Built context with {len(messages)} total messages (includes history + user context)")
 
     # Ensure session exists with metadata and eval mode if applicable
@@ -533,9 +565,17 @@ async def chat_completions(body: ChatCompletionRequest, request: Request):
         model_override=body.model,  # type: ignore[arg-type]
     )
 
-    # Combine all messages into single prompt for agent
-    # ContextBuilder already assembled: system context + history + new messages
-    prompt = "\n".join(msg.content for msg in messages)
+    # Build the prompt for the agent
+    # If we have proper message_history, use just the latest user message as prompt
+    # Otherwise, fall back to concatenating all messages (legacy behavior)
+    if pydantic_message_history:
+        # Use the latest user message as the prompt, with history passed separately
+        user_prompt = body.messages[-1].content if body.messages else ""
+        prompt = user_prompt
+        logger.debug(f"Using message_history with {len(pydantic_message_history)} messages")
+    else:
+        # Legacy: Combine all messages into single prompt for agent
+        prompt = "\n".join(msg.content for msg in messages)
 
     # Generate OpenAI-compatible request ID
     request_id = f"chatcmpl-{uuid.uuid4().hex[:24]}"
@@ -570,6 +610,8 @@ async def chat_completions(body: ChatCompletionRequest, request: Request):
                 agent_schema=schema_name,
                 session_id=context.session_id,
                 user_id=context.user_id,
+                agent_context=context,  # Pass context for multi-agent support
+                message_history=pydantic_message_history,  # Native pydantic-ai message history
             ),
             media_type="text/event-stream",
             headers={"Cache-Control": "no-cache", "Connection": "keep-alive"},
@@ -592,10 +634,16 @@ async def chat_completions(body: ChatCompletionRequest, request: Request):
         ) as span:
             # Capture trace context from the span we just created
             trace_id, span_id = get_current_trace_context()
-            result = await agent.run(prompt)
+            if pydantic_message_history:
+                result = await agent.run(prompt, message_history=pydantic_message_history)
+            else:
+                result = await agent.run(prompt)
     else:
         # No tracer available, run without tracing
-        result = await agent.run(prompt)
+        if pydantic_message_history:
+            result = await agent.run(prompt, message_history=pydantic_message_history)
+        else:
+            result = await agent.run(prompt)
 
     # Determine content format based on response_format request
     if body.response_format and body.response_format.type == "json_object":

rem/api/routers/chat/sse_events.py

@@ -321,7 +321,11 @@ class MetadataEvent(BaseModel):
     # Agent info
     agent_schema: str | None = Field(
         default=None,
-        description="Name of the agent schema used for this response (e.g., 'rem', 'query-assistant')"
+        description="Name of the top-level agent schema (e.g., 'siggy', 'rem')"
+    )
+    responding_agent: str | None = Field(
+        default=None,
+        description="Name of the agent that produced this response (may differ from agent_schema if delegated via ask_agent)"
     )
 
     # Session info
@@ -409,9 +413,9 @@ class ToolCallEvent(BaseModel):
         default=None,
         description="Tool arguments (for 'started' status)"
     )
-    result: str | None = Field(
+    result: str | dict[str, Any] | None = Field(
         default=None,
-        description="Tool result summary (for 'completed' status)"
+        description="Tool result - full dict for finalize_intake, summary string for others"
     )
     error: str | None = Field(
         default=None,