remdb 0.3.226__py3-none-any.whl → 0.3.245__py3-none-any.whl

rem/agentic/README.md

@@ -718,263 +718,37 @@ See `rem/api/README.md` for full SSE event protocol documentation.
 
 ## Multi-Agent Orchestration
 
-REM supports hierarchical agent orchestration where agents can delegate work to other agents via the `ask_agent` tool. This enables complex workflows with specialized agents.
-
-### Architecture
-
-```mermaid
-sequenceDiagram
-    participant User
-    participant API as Chat API
-    participant Orchestrator as Orchestrator Agent
-    participant EventSink as Event Sink (Queue)
-    participant Child as Child Agent
-    participant DB as PostgreSQL
-
-    User->>API: POST /chat/completions (stream=true)
-    API->>API: Create event sink (asyncio.Queue)
-    API->>Orchestrator: agent.iter(prompt)
-
-    loop Streaming Loop
-        Orchestrator->>API: PartDeltaEvent (text)
-        API->>User: SSE: data: {"delta": {"content": "..."}}
-    end
-
-    Orchestrator->>Orchestrator: Decides to call ask_agent
-    Orchestrator->>API: ToolCallPart (ask_agent)
-    API->>User: SSE: event: tool_call
-
-    API->>Child: ask_agent("child_name", input)
-    Child->>EventSink: push_event(child_content)
-    EventSink->>API: Consume child events
-    API->>User: SSE: data: {"delta": {"content": "..."}}
-
-    Child->>Child: Completes
-    Child-->>Orchestrator: Return result
-
-    Orchestrator->>API: Final response
-    API->>DB: Save tool calls
-    API->>DB: Save assistant message
-    API->>User: SSE: data: [DONE]
-```
-
-### Event Sink Pattern
-
-When an agent delegates to a child via `ask_agent`, the child's streaming events need to bubble up to the parent's stream. This is achieved through an **event sink** pattern using Python's `ContextVar`:
-
-```python
-# context.py
-from contextvars import ContextVar
-
-_parent_event_sink: ContextVar["asyncio.Queue | None"] = ContextVar(
-    "parent_event_sink", default=None
-)
-
-async def push_event(event: Any) -> bool:
-    """Push event to parent's event sink if available."""
-    sink = _parent_event_sink.get()
-    if sink is not None:
-        await sink.put(event)
-        return True
-    return False
-```
-
-The streaming controller sets up the event sink before agent execution:
-
-```python
-# streaming.py
-child_event_sink: asyncio.Queue = asyncio.Queue()
-set_event_sink(child_event_sink)
-
-async for node in agent.iter(prompt):
-    # Process agent events...
-
-    # Consume any child events that arrived
-    while not child_event_sink.empty():
-        child_event = child_event_sink.get_nowait()
-        if child_event["type"] == "child_content":
-            yield format_sse_content_delta(child_event["content"])
-```
-
-### ask_agent Tool Implementation
-
-The `ask_agent` tool in `mcp_router/tools.py` uses Pydantic AI's streaming iteration:
-
-```python
-async def ask_agent(agent_name: str, input_text: str, ...):
-    """Delegate work to another agent."""
-
-    # Load and create child agent
-    schema = await load_agent_schema_async(agent_name, user_id)
-    child_agent = await create_agent(context=context, agent_schema_override=schema)
-
-    # Stream child agent with event proxying
-    async with child_agent.iter(prompt) as agent_run:
-        async for node in agent_run:
-            if Agent.is_model_request_node(node):
-                async with node.stream(agent_run.ctx) as request_stream:
-                    async for event in request_stream:
-                        if isinstance(event, PartDeltaEvent):
-                            # Push content to parent's event sink
-                            await push_event({
-                                "type": "child_content",
-                                "agent_name": agent_name,
-                                "content": event.delta.content_delta,
-                            })
-
-    return agent_run.result
-```
-
-### Pydantic AI Features Used
+Agents can delegate work to other agents via the `ask_agent` tool. This enables orchestrator patterns where a parent agent routes to specialists.
 
-#### 1. Streaming Iteration (`agent.iter()`)
+### How It Works
 
-Unlike `agent.run()` which blocks until completion, `agent.iter()` provides fine-grained control over the execution flow:
+1. **Parent agent** calls `ask_agent(agent_name, input_text)`
+2. **Child agent** executes and streams its response
+3. **Child events** bubble up to parent via an event sink (asyncio.Queue in ContextVar)
+4. **All tool calls** are saved to the database for the session
 
-```python
-async with agent.iter(prompt) as agent_run:
-    async for node in agent_run:
-        if Agent.is_model_request_node(node):
-            # Model is generating - stream the response
-            async with node.stream(agent_run.ctx) as stream:
-                async for event in stream:
-                    if isinstance(event, PartStartEvent):
-                        # Tool call starting
-                    elif isinstance(event, PartDeltaEvent):
-                        # Content chunk
-        elif Agent.is_call_tools_node(node):
-            # Tools are being executed
-            async with node.stream(agent_run.ctx) as stream:
-                async for event in stream:
-                    if isinstance(event, FunctionToolResultEvent):
-                        # Tool completed
-```
-
-#### 2. Node Types
-
-- **`ModelRequestNode`**: The model is generating a response (text or tool calls)
-- **`CallToolsNode`**: Tools are being executed
-- **`End`**: Agent execution complete
-
-#### 3. Event Types
+### Key Components
 
-- **`PartStartEvent`**: A new part (text or tool call) is starting
-- **`PartDeltaEvent`**: Content chunk for streaming text
-- **`FunctionToolResultEvent`**: Tool execution completed with result
-- **`ToolCallPart`**: Metadata about a tool call (name, arguments)
-- **`TextPart`**: Text content
+| Component | Location | Purpose |
+|-----------|----------|---------|
+| `ask_agent` tool | `mcp_router/tools.py` | Loads child agent, runs with streaming, pushes events to sink |
+| Event sink | `context.py` | ContextVar holding asyncio.Queue for child→parent event flow |
+| Streaming controller | `streaming.py` | Drains event sink, emits SSE events, saves to DB |
 
-### Message Persistence
-
-All messages are persisted to PostgreSQL for session continuity:
-
-```python
-# streaming.py - after agent completes
-async def save_session_messages(...):
-    store = SessionMessageStore(user_id=user_id)
-
-    # Save each tool call as a tool message
-    for tool_call in tool_calls:
-        await store.save_message(
-            session_id=session_id,
-            role="tool",
-            content=tool_call.result,
-            tool_name=tool_call.name,
-            tool_call_id=tool_call.id,
-        )
-
-    # Save the final assistant response
-    await store.save_message(
-        session_id=session_id,
-        role="assistant",
-        content=accumulated_content,
-    )
-```
+### Event Types
 
-Messages are stored with:
-- **Embeddings**: For semantic search across conversation history
-- **Compression**: Long conversations are summarized to manage context window
-- **Session isolation**: Each session maintains its own message history
+Child agents emit events that the parent streams to the client:
 
-### Testing Multi-Agent Systems
+- **`child_tool_start`**: Child is calling a tool (logged, streamed, saved to DB)
+- **`child_content`**: Child's text response (streamed as SSE content delta)
+- **`child_tool_result`**: Tool completed with result (metadata extraction)
 
-#### Integration Tests
+### Testing
 
-Real end-to-end tests without mocking are in `tests/integration/test_ask_agent_streaming.py`:
-
-```python
-class TestAskAgentStreaming:
-    async def test_ask_agent_streams_and_saves(self, session_id, user_id):
-        """Test delegation via ask_agent."""
-        # Uses test_orchestrator which always delegates to test_responder
-        agent = await create_agent(context=context, agent_schema_override=schema)
-
-        chunks = []
-        async for chunk in stream_openai_response_with_save(
-            agent=agent,
-            prompt="Hello, please delegate this",
-            ...
-        ):
-            chunks.append(chunk)
-
-        # Verify streaming worked
-        assert len(content_chunks) > 0
-
-        # Verify persistence
-        messages = await store.load_session_messages(session_id)
-        assert len([m for m in messages if m["role"] == "assistant"]) == 1
-        assert len([m for m in messages if m["tool_name"] == "ask_agent"]) >= 1
-
-    async def test_multi_turn_saves_all_assistant_messages(self, session_id, user_id):
-        """Test that each turn saves its own assistant message.
-
-        This catches scoping bugs like accumulated_content not being
-        properly scoped per-turn.
-        """
-        turn_prompts = [
-            "Hello, how are you?",
-            "Tell me something interesting",
-            "Thanks for chatting!",
-        ]
-
-        for prompt in turn_prompts:
-            async for chunk in stream_openai_response_with_save(...):
-                pass
-
-        # Each turn should save an assistant message
-        messages = await store.load_session_messages(session_id)
-        assistant_msgs = [m for m in messages if m["role"] == "assistant"]
-        assert len(assistant_msgs) == 3
-```
-
-#### Test Agent Schemas
-
-Test agents are defined in `tests/data/schemas/agents/`:
-
-- **`test_orchestrator.yaml`**: Always delegates via `ask_agent`
-- **`test_responder.yaml`**: Simple agent that responds directly
-
-```yaml
-# test_orchestrator.yaml
-type: object
-description: |
-  You are a TEST ORCHESTRATOR that ALWAYS delegates to another agent.
-  Call ask_agent with agent_name="test_responder" on EVERY turn.
-json_schema_extra:
-  kind: agent
-  name: test_orchestrator
-  tools:
-    - name: ask_agent
-      mcp_server: rem
-```
-
-#### Running Integration Tests
-
-```bash
-# Run individually (recommended due to async isolation)
-POSTGRES__CONNECTION_STRING="postgresql://rem:rem@localhost:5050/rem" \
-  uv run pytest tests/integration/test_ask_agent_streaming.py::TestAskAgentStreaming::test_multi_turn_saves_all_assistant_messages -v -s
-```
+Integration tests in `tests/integration/test_ask_agent_streaming.py` verify:
+- Child content streams correctly to client
+- Tool calls are persisted to database
+- Multi-turn conversations save all messages
 
 ## Future Work
 

rem/agentic/context.py

@@ -16,6 +16,7 @@ Headers Mapping:
     X-Agent-Schema   → context.agent_schema_uri (default: "rem")
     X-Model-Name     → context.default_model
     X-Is-Eval        → context.is_eval (marks session as evaluation)
+    X-Client-Id      → context.client_id (e.g., "web", "mobile", "cli")
 
 Key Design Pattern:
 - AgentContext is passed to agent factory, not stored in agents
@@ -222,6 +223,11 @@ class AgentContext(BaseModel):
         description="Whether this is an evaluation session (set via X-Is-Eval header)",
     )
 
+    client_id: str | None = Field(
+        default=None,
+        description="Client identifier (e.g., 'web', 'mobile', 'cli') set via X-Client-Id header",
+    )
+
     model_config = {"populate_by_name": True}
 
     def child_context(
@@ -232,7 +238,7 @@ class AgentContext(BaseModel):
         """
         Create a child context for nested agent calls.
 
-        Inherits user_id, tenant_id, session_id, is_eval from parent.
+        Inherits user_id, tenant_id, session_id, is_eval, client_id from parent.
         Allows overriding agent_schema_uri and default_model for the child.
 
         Args:
@@ -256,6 +262,7 @@ class AgentContext(BaseModel):
             default_model=model_override or self.default_model,
             agent_schema_uri=agent_schema_uri or self.agent_schema_uri,
             is_eval=self.is_eval,
+            client_id=self.client_id,
         )
 
     @staticmethod
@@ -374,6 +381,7 @@ class AgentContext(BaseModel):
             default_model=normalized.get("x-model-name") or settings.llm.default_model,
             agent_schema_uri=normalized.get("x-agent-schema"),
             is_eval=is_eval,
+            client_id=normalized.get("x-client-id"),
         )
 
     @classmethod
@@ -391,6 +399,7 @@ class AgentContext(BaseModel):
         - X-Model-Name: Model override
         - X-Agent-Schema: Agent schema URI
         - X-Is-Eval: Whether this is an evaluation session (true/false)
+        - X-Client-Id: Client identifier (e.g., "web", "mobile", "cli")
 
         Args:
             headers: Dictionary of HTTP headers (case-insensitive)
@@ -404,7 +413,8 @@ class AgentContext(BaseModel):
                 "X-Tenant-Id": "acme-corp",
                 "X-Session-Id": "sess-456",
                 "X-Model-Name": "anthropic:claude-opus-4-20250514",
-                "X-Is-Eval": "true"
+                "X-Is-Eval": "true",
+                "X-Client-Id": "web"
             }
             context = AgentContext.from_headers(headers)
         """
@@ -422,4 +432,5 @@ class AgentContext(BaseModel):
             default_model=normalized.get("x-model-name") or settings.llm.default_model,
             agent_schema_uri=normalized.get("x-agent-schema"),
             is_eval=is_eval,
+            client_id=normalized.get("x-client-id"),
         )

rem/agentic/context_builder.py

@@ -4,15 +4,12 @@ Centralized context builder for agent execution.
 Session History (ALWAYS loaded with compression):
 - Each chat request is a single message, so session history MUST be recovered
 - Uses SessionMessageStore with compression to keep context efficient
-- Long assistant responses include REM LOOKUP hints: "... [REM LOOKUP session-{id}-msg-{index}] ..."
-- Agent can retrieve full content on-demand using REM LOOKUP
 - Prevents context window bloat while maintaining conversation continuity
 
 User Context (on-demand by default):
-- System message includes REM LOOKUP hint for user profile
-- Agent decides whether to load profile based on query
-- More efficient for queries that don't need personalization
-- Example: "User: sarah@example.com. To load user profile: Use REM LOOKUP \"sarah@example.com\""
+- System message includes user email for context awareness
+- Fails silently if user not found - agent proceeds without user context
+- Example: "User: sarah@example.com"
 
 User Context (auto-inject when enabled):
 - Set CHAT__AUTO_INJECT_USER_CONTEXT=true
@@ -22,8 +19,8 @@ User Context (auto-inject when enabled):
 Design Pattern:
 1. Extract AgentContext from headers (user_id, tenant_id, session_id)
 2. If auto-inject enabled: Load User/Session from database
-3. If auto-inject disabled: Provide REM LOOKUP hints in system message
-4. Construct system message with date + context (injected or hints)
+3. If auto-inject disabled: Show user email for context (fail silently if not found)
+4. Construct system message with date + context
 5. Return complete context ready for agent execution
 
 Integration Points:
@@ -40,11 +37,10 @@ Usage (on-demand, default):
 
     # Messages list structure (on-demand):
     # [
-    #   {"role": "system", "content": "Today's date: 2025-11-22\nUser: sarah@example.com\nTo load user profile: Use REM LOOKUP \"sarah@example.com\"\nSession ID: sess-123\nTo load session history: Use REM LOOKUP messages?session_id=sess-123"},
+    #   {"role": "system", "content": "Today's date: 2025-11-22\n\nUser: sarah@example.com"},
     #   {"role": "user", "content": "What's next for the API migration?"}
     # ]
 
-    # Agent receives hints and can decide to load context if needed
     agent = await create_agent(context=context, ...)
     prompt = "\n".join(msg.content for msg in messages)
     result = await agent.run(prompt)
@@ -52,7 +48,7 @@ Usage (on-demand, default):
 Usage (auto-inject, CHAT__AUTO_INJECT_USER_CONTEXT=true):
     # Messages list structure (auto-inject):
     # [
-    #   {"role": "system", "content": "Today's date: 2025-11-22\n\nUser Context (auto-injected):\nSummary: ...\nInterests: ...\n\nSession History (auto-injected, 5 messages):"},
+    #   {"role": "system", "content": "Today's date: 2025-11-22\n\nUser Context (auto-injected):\nSummary: ...\nInterests: ..."},
     #   {"role": "user", "content": "Previous message"},
     #   {"role": "assistant", "content": "Previous response"},
     #   {"role": "user", "content": "What's next for the API migration?"}
@@ -110,13 +106,11 @@ class ContextBuilder:
 
         Session History (ALWAYS loaded with compression):
         - If session_id provided, session history is ALWAYS loaded using SessionMessageStore
-        - Compression keeps it efficient with REM LOOKUP hints for long messages
-        - Example: "... [Message truncated - REM LOOKUP session-{id}-msg-{index}] ..."
-        - Agent can retrieve full content on-demand using REM LOOKUP
+        - Compression keeps context efficient
 
         User Context (on-demand by default):
-        - System message includes REM LOOKUP hint: "User: {email}. To load user profile: Use REM LOOKUP \"{email}\""
-        - Agent decides whether to load profile based on query
+        - System message includes user email: "User: {email}"
+        - Fails silently if user not found - agent proceeds without user context
 
         User Context (auto-inject when enabled):
         - Set CHAT__AUTO_INJECT_USER_CONTEXT=true
@@ -137,9 +131,9 @@ class ContextBuilder:
 
             # messages structure:
             # [
-            #   {"role": "system", "content": "Today's date: 2025-11-22\nUser: sarah@example.com\nTo load user profile: Use REM LOOKUP \"sarah@example.com\""},
+            #   {"role": "system", "content": "Today's date: 2025-11-22\n\nUser: sarah@example.com"},
             #   {"role": "user", "content": "Previous message"},
-            #   {"role": "assistant", "content": "Start of long response... [REM LOOKUP session-123-msg-1] ...end"},
+            #   {"role": "assistant", "content": "Previous response"},
             #   {"role": "user", "content": "New message"}
             # ]
         """
@@ -158,6 +152,7 @@ class ContextBuilder:
                 default_model=context.default_model,
                 agent_schema_uri=context.agent_schema_uri,
                 is_eval=context.is_eval,
+                client_id=context.client_id,
             )
 
         # Initialize DB if not provided and needed (for user context or session history)
@@ -177,6 +172,10 @@ class ContextBuilder:
             today = datetime.now(timezone.utc).strftime("%Y-%m-%d")
             context_hint = f"Today's date: {today}."
 
+            # Add client identifier if present
+            if context.client_id:
+                context_hint += f"\nClient: {context.client_id}"
+
             # Add user context (auto-inject or on-demand hint)
             if settings.chat.auto_inject_user_context and context.user_id and db:
                 # Auto-inject: Load and include user profile
@@ -189,18 +188,18 @@ class ContextBuilder:
                     context_hint += f"\n\nUser Context (auto-injected):\n{user_context_content}"
                 else:
                     context_hint += "\n\nNo user context available (anonymous or new user)."
-            elif context.user_id:
-                # On-demand: Provide hint to use REM LOOKUP
-                # user_id is UUID5 hash of email - load user to get email for display and LOOKUP
-                user_repo = Repository(User, "users", db=db)
-                user = await user_repo.get_by_id(context.user_id, context.tenant_id)
-                if user and user.email:
-                    # Show email (more useful than UUID) and LOOKUP hint
-                    context_hint += f"\n\nUser: {user.email}"
-                    context_hint += f"\nTo load user profile: Use REM LOOKUP \"{user.email}\""
-                else:
-                    context_hint += f"\n\nUser ID: {context.user_id}"
-                    context_hint += "\nUser profile not available."
+            elif context.user_id and db:
+                # On-demand: Show user email for context (no REM LOOKUP - it requires exact user_id match)
+                # Fail silently if user lookup fails - just proceed without user context
+                try:
+                    user_repo = Repository(User, "users", db=db)
+                    user = await user_repo.get_by_id(context.user_id, context.tenant_id)
+                    if user and user.email:
+                        context_hint += f"\n\nUser: {user.email}"
+                    # If user not found, just proceed without adding user context
+                except Exception as e:
+                    # Fail silently - don't block agent execution if user lookup fails
+                    logger.debug(f"Could not load user context: {e}")
 
             # Add system context hint
             messages.append(ContextMessage(role="system", content=context_hint))
@@ -318,6 +317,7 @@ class ContextBuilder:
         session_id: str | None = None,
         message: str = "Hello",
         model: str | None = None,
+        client_id: str | None = None,
     ) -> tuple[AgentContext, list[ContextMessage]]:
         """
         Build context for testing (no database lookup).
@@ -325,7 +325,7 @@ class ContextBuilder:
         Creates minimal context with:
         - Test user (test@rem.ai)
         - Test tenant
-        - Context hint with date
+        - Context hint with date and client
         - Single user message
 
         Args:
@@ -334,6 +334,7 @@ class ContextBuilder:
             session_id: Optional session ID
             message: User message content
             model: Optional model override
+            client_id: Optional client identifier (e.g., "cli", "test")
 
         Returns:
             Tuple of (AgentContext, messages list)
@@ -341,7 +342,8 @@ class ContextBuilder:
         Example:
             context, messages = await ContextBuilder.build_from_test(
                 user_id="test@rem.ai",
-                message="What's the weather like?"
+                message="What's the weather like?",
+                client_id="cli"
             )
         """
         from ..settings import settings
@@ -352,11 +354,15 @@ class ContextBuilder:
             tenant_id=tenant_id,
             session_id=session_id,
             default_model=model or settings.llm.default_model,
+            client_id=client_id,
         )
 
         # Build minimal messages
         today = datetime.now(timezone.utc).strftime("%Y-%m-%d")
-        context_hint = f"Today's date: {today}.\n\nTest user context: {user_id} (test mode, no profile loaded)."
+        context_hint = f"Today's date: {today}."
+        if client_id:
+            context_hint += f"\nClient: {client_id}"
+        context_hint += f"\n\nTest user context: {user_id} (test mode, no profile loaded)."
 
         messages = [
             ContextMessage(role="system", content=context_hint),