remdb 0.3.146__py3-none-any.whl → 0.3.181__py3-none-any.whl

rem/api/routers/chat/completions.py

@@ -97,7 +97,7 @@ Context Building Flow:
    - Long messages include REM LOOKUP hints: "... [REM LOOKUP session-{id}-msg-{index}] ..."
    - Agent can retrieve full content on-demand using REM LOOKUP
 3. User profile provided as REM LOOKUP hint (on-demand by default)
-   - Agent receives: "User ID: {user_id}. To load user profile: Use REM LOOKUP users/{user_id}"
+   - Agent receives: "User: {email}. To load user profile: Use REM LOOKUP \"{email}\""
    - Agent decides whether to load profile based on query
 4. If CHAT__AUTO_INJECT_USER_CONTEXT=true: User profile auto-loaded and injected
 5. Combines: system context + compressed session history + new messages
@@ -330,8 +330,8 @@ async def chat_completions(body: ChatCompletionRequest, request: Request):
     - Useful for A/B testing, model comparison, and feedback collection
     """
     # Load agent schema: use header value from context or default
-    # Extract AgentContext first to get schema name
-    temp_context = AgentContext.from_headers(dict(request.headers))
+    # Extract AgentContext from request (gets user_id from JWT token)
+    temp_context = AgentContext.from_request(request)
     schema_name = temp_context.agent_schema_uri or DEFAULT_AGENT_SCHEMA
 
     # Resolve model: use body.model if provided, otherwise settings default
@@ -350,6 +350,7 @@ async def chat_completions(body: ChatCompletionRequest, request: Request):
         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)
         )
 
         # Ensure session exists with metadata and eval mode if applicable
@@ -509,6 +510,7 @@ async def chat_completions(body: ChatCompletionRequest, request: Request):
     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)
     )
 
     logger.info(f"Built context with {len(messages)} total messages (includes history + user context)")

rem/api/routers/chat/streaming.py

@@ -76,6 +76,9 @@ async def stream_openai_response(
     agent_schema: str | None = None,
     # Mutable container to capture trace context (deterministic, not AI-dependent)
     trace_context_out: dict | None = None,
+    # Mutable container to capture tool calls for persistence
+    # Format: list of {"tool_name": str, "tool_id": str, "arguments": dict, "result": any}
+    tool_calls_out: list | None = None,
 ) -> AsyncGenerator[str, None]:
     """
     Stream Pydantic AI agent responses with rich SSE events.
@@ -146,6 +149,9 @@ async def stream_openai_response(
     pending_tool_completions: list[tuple[str, str]] = []
     # Track if metadata was registered via register_metadata tool
     metadata_registered = False
+    # Track pending tool calls with full data for persistence
+    # Maps tool_id -> {"tool_name": str, "tool_id": str, "arguments": dict}
+    pending_tool_data: dict[str, dict] = {}
 
     try:
         # Emit initial progress event
@@ -299,6 +305,13 @@ async def stream_openai_response(
                                     arguments=args_dict
                                 ))
 
+                                # Track tool call data for persistence (especially register_metadata)
+                                pending_tool_data[tool_id] = {
+                                    "tool_name": tool_name,
+                                    "tool_id": tool_id,
+                                    "arguments": args_dict,
+                                }
+
                                 # Update progress
                                 current_step = 2
                                 total_steps = 4  # Added tool execution step
@@ -421,6 +434,15 @@ async def stream_openai_response(
                                         hidden=False,
                                     ))
 
+                                # Capture tool call with result for persistence
+                                # Special handling for register_metadata - always capture full data
+                                if tool_calls_out is not None and tool_id in pending_tool_data:
+                                    tool_data = pending_tool_data[tool_id]
+                                    tool_data["result"] = result_content
+                                    tool_data["is_metadata"] = is_metadata_event
+                                    tool_calls_out.append(tool_data)
+                                    del pending_tool_data[tool_id]
+
                                 if not is_metadata_event:
                                     # Normal tool completion - emit ToolCallEvent
                                     result_str = str(result_content)
@@ -728,6 +750,9 @@ async def stream_openai_response_with_save(
     # Accumulate content during streaming
     accumulated_content = []
 
+    # Capture tool calls for persistence (especially register_metadata)
+    tool_calls: list = []
+
     async for chunk in stream_openai_response(
         agent=agent,
         prompt=prompt,
@@ -737,6 +762,7 @@ async def stream_openai_response_with_save(
         session_id=session_id,
         message_id=message_id,
         trace_context_out=trace_context,  # Pass container to capture trace IDs
+        tool_calls_out=tool_calls,  # Capture tool calls for persistence
     ):
         yield chunk
 
@@ -755,28 +781,75 @@ async def stream_openai_response_with_save(
             except (json.JSONDecodeError, KeyError, IndexError):
                 pass  # Skip non-JSON or malformed chunks
 
-    # After streaming completes, save the assistant response
-    if settings.postgres.enabled and session_id and accumulated_content:
-        full_content = "".join(accumulated_content)
+    # After streaming completes, save tool calls and assistant response
+    # Note: All messages stored UNCOMPRESSED. Compression happens on reload.
+    if settings.postgres.enabled and session_id:
         # Get captured trace context from container (deterministically captured inside agent execution)
         captured_trace_id = trace_context.get("trace_id")
         captured_span_id = trace_context.get("span_id")
-        assistant_message = {
-            "id": message_id,  # Use pre-generated ID for consistency with metadata event
-            "role": "assistant",
-            "content": full_content,
-            "timestamp": to_iso(utc_now()),
-            "trace_id": captured_trace_id,
-            "span_id": captured_span_id,
-        }
-        try:
-            store = SessionMessageStore(user_id=user_id or settings.test.effective_user_id)
-            await store.store_session_messages(
-                session_id=session_id,
-                messages=[assistant_message],
-                user_id=user_id,
-                compress=True,  # Compress long assistant responses
-            )
-            logger.debug(f"Saved assistant response {message_id} to session {session_id} ({len(full_content)} chars)")
-        except Exception as e:
-            logger.error(f"Failed to save assistant response: {e}", exc_info=True)
+        timestamp = to_iso(utc_now())
+
+        messages_to_store = []
+
+        # First, store tool call messages (message_type: "tool")
+        for tool_call in tool_calls:
+            tool_message = {
+                "role": "tool",
+                "content": json.dumps(tool_call.get("result", {}), default=str),
+                "timestamp": timestamp,
+                "trace_id": captured_trace_id,
+                "span_id": captured_span_id,
+                # Store tool call details in a way that can be reconstructed
+                "tool_call_id": tool_call.get("tool_id"),
+                "tool_name": tool_call.get("tool_name"),
+                "tool_arguments": tool_call.get("arguments"),
+            }
+            messages_to_store.append(tool_message)
+
+        # Then store assistant text response (if any)
+        if accumulated_content:
+            full_content = "".join(accumulated_content)
+            assistant_message = {
+                "id": message_id,  # Use pre-generated ID for consistency with metadata event
+                "role": "assistant",
+                "content": full_content,
+                "timestamp": timestamp,
+                "trace_id": captured_trace_id,
+                "span_id": captured_span_id,
+            }
+            messages_to_store.append(assistant_message)
+
+        if messages_to_store:
+            try:
+                store = SessionMessageStore(user_id=user_id or settings.test.effective_user_id)
+                await store.store_session_messages(
+                    session_id=session_id,
+                    messages=messages_to_store,
+                    user_id=user_id,
+                    compress=False,  # Store uncompressed; compression happens on reload
+                )
+                logger.debug(
+                    f"Saved {len(tool_calls)} tool calls and "
+                    f"{'assistant response' if accumulated_content else 'no text'} "
+                    f"to session {session_id}"
+                )
+            except Exception as e:
+                logger.error(f"Failed to save session messages: {e}", exc_info=True)
+
+        # Update session description with session_name (non-blocking, after all yields)
+        for tool_call in tool_calls:
+            if tool_call.get("tool_name") == "register_metadata" and tool_call.get("is_metadata"):
+                session_name = tool_call.get("arguments", {}).get("session_name")
+                if session_name:
+                    try:
+                        from ....models.entities import Session
+                        from ....services.postgres import Repository
+                        repo = Repository(Session, table_name="sessions")
+                        session = await repo.get_by_id(session_id)
+                        if session and session.description != session_name:
+                            session.description = session_name
+                            await repo.update(session)
+                            logger.debug(f"Updated session {session_id} description to '{session_name}'")
+                    except Exception as e:
+                        logger.warning(f"Failed to update session description: {e}")
+                    break

rem/api/routers/messages.py

@@ -134,7 +134,6 @@ async def list_messages(
     ),
     limit: int = Query(default=50, ge=1, le=100, description="Max results to return"),
     offset: int = Query(default=0, ge=0, description="Offset for pagination"),
-    x_tenant_id: str = Header(alias="X-Tenant-Id", default="default"),
 ) -> MessageListResponse:
     """
     List messages with optional filters.
@@ -158,15 +157,18 @@ async def list_messages(
 
     repo = Repository(Message, table_name="messages")
 
+    # Get current user for logging
+    current_user = get_current_user(request)
+    jwt_user_id = current_user.get("id") if current_user else None
+
     # If mine=true, force filter to current user's ID from JWT
     effective_user_id = user_id
     if mine:
-        current_user = get_current_user(request)
         if current_user:
             effective_user_id = current_user.get("id")
 
     # Build user-scoped filters (admin can see all, regular users see only their own)
-    filters = await get_user_filter(request, x_user_id=effective_user_id, x_tenant_id=x_tenant_id)
+    filters = await get_user_filter(request, x_user_id=effective_user_id)
 
     # Apply optional filters
     if session_id:
@@ -174,6 +176,13 @@ async def list_messages(
     if message_type:
         filters["message_type"] = message_type
 
+    # Log the query parameters for debugging
+    logger.debug(
+        f"[messages] Query: session_id={session_id} | "
+        f"jwt_user_id={jwt_user_id} | "
+        f"filters={filters}"
+    )
+
     # For date filtering, we need custom SQL (not supported by basic Repository)
     # For now, fetch all matching base filters and filter in Python
     # TODO: Extend Repository to support date range filters
@@ -206,6 +215,12 @@ async def list_messages(
     # Get total count for pagination info
     total = await repo.count(filters)
 
+    # Log result count
+    logger.debug(
+        f"[messages] Result: returned={len(messages)} | total={total} | "
+        f"session_id={session_id}"
+    )
+
     return MessageListResponse(data=messages, total=total, has_more=has_more)
 
 
@@ -213,7 +228,6 @@ async def list_messages(
 async def get_message(
     request: Request,
     message_id: str,
-    x_tenant_id: str = Header(alias="X-Tenant-Id", default="default"),
 ) -> Message:
     """
     Get a specific message by ID.
@@ -236,7 +250,7 @@ async def get_message(
         raise HTTPException(status_code=503, detail="Database not enabled")
 
     repo = Repository(Message, table_name="messages")
-    message = await repo.get_by_id(message_id, x_tenant_id)
+    message = await repo.get_by_id(message_id)
 
     if not message:
         raise HTTPException(status_code=404, detail=f"Message '{message_id}' not found")
@@ -263,7 +277,6 @@ async def list_sessions(
     mode: SessionMode | None = Query(default=None, description="Filter by session mode"),
     page: int = Query(default=1, ge=1, description="Page number (1-indexed)"),
     page_size: int = Query(default=50, ge=1, le=100, description="Number of results per page"),
-    x_tenant_id: str = Header(alias="X-Tenant-Id", default="default"),
 ) -> SessionsQueryResponse:
     """
     List sessions with optional filters and page-based pagination.
@@ -288,7 +301,7 @@ async def list_sessions(
     repo = Repository(Session, table_name="sessions")
 
     # Build user-scoped filters (admin can see all, regular users see only their own)
-    filters = await get_user_filter(request, x_user_id=user_id, x_tenant_id=x_tenant_id)
+    filters = await get_user_filter(request, x_user_id=user_id)
     if mode:
         filters["mode"] = mode.value
 
@@ -319,7 +332,6 @@ async def create_session(
     request_body: SessionCreateRequest,
     user: dict = Depends(require_admin),
     x_user_id: str = Header(alias="X-User-Id", default="default"),
-    x_tenant_id: str = Header(alias="X-Tenant-Id", default="default"),
 ) -> Session:
     """
     Create a new session.
@@ -334,7 +346,6 @@ async def create_session(
 
     Headers:
     - X-User-Id: User identifier (owner of the session)
-    - X-Tenant-Id: Tenant identifier
 
     Returns:
         Created session object
@@ -354,7 +365,7 @@ async def create_session(
         prompt=request_body.prompt,
         agent_schema_uri=request_body.agent_schema_uri,
         user_id=effective_user_id,
-        tenant_id=x_tenant_id,
+        tenant_id="default",  # tenant_id not used for filtering, set to default
     )
 
     repo = Repository(Session, table_name="sessions")
@@ -372,7 +383,6 @@ async def create_session(
 async def get_session(
     request: Request,
     session_id: str,
-    x_tenant_id: str = Header(alias="X-Tenant-Id", default="default"),
 ) -> Session:
     """
     Get a specific session by ID.
@@ -395,11 +405,11 @@ async def get_session(
         raise HTTPException(status_code=503, detail="Database not enabled")
 
     repo = Repository(Session, table_name="sessions")
-    session = await repo.get_by_id(session_id, x_tenant_id)
+    session = await repo.get_by_id(session_id)
 
     if not session:
         # Try finding by name
-        sessions = await repo.find({"name": session_id, "tenant_id": x_tenant_id}, limit=1)
+        sessions = await repo.find({"name": session_id}, limit=1)
         if sessions:
             session = sessions[0]
         else:
@@ -420,7 +430,6 @@ async def update_session(
     request: Request,
     session_id: str,
     request_body: SessionUpdateRequest,
-    x_tenant_id: str = Header(alias="X-Tenant-Id", default="default"),
 ) -> Session:
     """
     Update an existing session.
@@ -450,7 +459,7 @@ async def update_session(
         raise HTTPException(status_code=503, detail="Database not enabled")
 
     repo = Repository(Session, table_name="sessions")
-    session = await repo.get_by_id(session_id, x_tenant_id)
+    session = await repo.get_by_id(session_id)
 
     if not session:
         raise HTTPException(status_code=404, detail=f"Session '{session_id}' not found")

rem/auth/__init__.py

@@ -1,26 +1,36 @@
 """
 REM Authentication Module.
 
-OAuth 2.1 compliant authentication with support for:
+Authentication with support for:
+- Email passwordless login (verification codes)
 - Google OAuth
 - Microsoft Entra ID (Azure AD) OIDC
 - Custom OIDC providers
 
 Design Pattern:
 - Provider-agnostic base classes
-- PKCE (Proof Key for Code Exchange) for all flows
+- PKCE (Proof Key for Code Exchange) for OAuth flows
 - State parameter for CSRF protection
 - Nonce for ID token replay protection
 - Token validation with JWKS
-- Clean separation: providers/ for OAuth logic, middleware.py for FastAPI integration
+- Clean separation: providers/ for auth logic, middleware.py for FastAPI integration
+
+Email Auth Flow:
+1. POST /api/auth/email/send-code with {email}
+2. User receives code via email
+3. POST /api/auth/email/verify with {email, code}
+4. Session created, user authenticated
 """
 
 from .providers.base import OAuthProvider
+from .providers.email import EmailAuthProvider, EmailAuthResult
 from .providers.google import GoogleOAuthProvider
 from .providers.microsoft import MicrosoftOAuthProvider
 
 __all__ = [
     "OAuthProvider",
+    "EmailAuthProvider",
+    "EmailAuthResult",
     "GoogleOAuthProvider",
     "MicrosoftOAuthProvider",
 ]