Add LLM token

backend/app/api/routes/agent.py

@@ -20,7 +20,11 @@ from app.api.models import (
 )
 from app.config.settings import settings
 from app.shared.async_utils import iter_in_thread
-from app.shared.bootstrap import get_agent_conversation_service, get_agent_session_service
+from app.shared.bootstrap import (
+    get_agent_conversation_service,
+    get_agent_session_service,
+    get_agentic_conversation_service,
+)
 # Keep route handlers close to their transport-layer wiring for easier auditing.
 
 
@@ -182,3 +186,58 @@ async def submit_feedback(request: FeedbackRequest):
         return {"message": "反馈已提交", "session_id": result.session_id, "message_index": result.message_index}
     except ValueError as exc:
         raise HTTPException(status_code=404, detail=str(exc))
+
+
+# ── P0-1: Agentic RAG endpoint ────────────────────────────────────────────────
+
+@router.post("/agentic/stream")
+async def agentic_stream(request: ChatRequest):
+    """Stream an Agentic RAG response with live multi-step reasoning trace.
+
+    Unlike the standard ``/chat/stream`` endpoint this route runs a full pipeline:
+    intent analysis → query planning → iterative retrieval → grounding check →
+    answer generation.
+
+    Extra SSE event types beyond the standard ones:
+
+    * ``thinking`` — reasoning sub-step progress; data is a JSON object with
+      ``step`` (intent_analysis / query_planning / retrieving / grounding_check),
+      ``status`` (running / done), and step-specific fields.
+
+    The ``sources``, ``content``, and ``done`` events are identical to the standard
+    chat-stream contract so the existing frontend parser can handle them without
+    changes.
+    """
+    async def generate_sse() -> AsyncGenerator[str, None]:
+        """Handle SSE generation for the agentic chat endpoint."""
+        try:
+            session_id_, event_stream = get_agentic_conversation_service().stream_agentic_chat(
+                query=request.query,
+                session_id=request.session_id,
+                filters=request.filters,
+                provider=request.provider or settings.llm_provider,
+                model=request.model or settings.llm_model,
+                top_k=request.top_k or settings.rag_top_k,
+                context_text=request.context_text,
+                context_filename=request.context_filename,
+            )
+            yield f"event: session\ndata: {json.dumps({'session_id': session_id_})}\n\n"
+            async for event_data in iter_in_thread(event_stream):
+                event_type = event_data.get("event", "content")
+                data = event_data.get("data", "")
+                if isinstance(data, (dict, list)):
+                    yield f"event: {event_type}\ndata: {json.dumps(data, ensure_ascii=False)}\n\n"
+                else:
+                    yield f"event: {event_type}\ndata: {data}\n\n"
+        except Exception as exc:
+            yield f"event: error\ndata: {str(exc)}\n\n"
+
+    return StreamingResponse(
+        generate_sse(),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+            "X-Accel-Buffering": "no",
+        },
+    )