Execute & Traces
Recipe chatbot (reference SUT)
Use the recipe-chatbot HTTP API as the canonical system under test for tutorials.
The recipe-chatbot is the reference system under test for QAVE Studio tutorials. It is a conversational FastAPI app with optional Langfuse tracing.
Endpoints
| Method | Path | Purpose |
|---|---|---|
POST | /api/chat | Primary SUT surface |
GET | /health | Readiness |
GET | / | Chat UI |
Chat contract
// Request
{ "session_id": "optional", "message": "..." }
// or OpenAI-style messages[]
// Response
{ "session_id": "...", "reply": "...", "model": "..." }Optional header: X-Session-Id.
Local ports
Compose often maps the app to 9000; avoid colliding with the Gateway on 8000.
curl -s http://localhost:9000/api/chat \
-H 'Content-Type: application/json' \
-d '{"message":"Give me a quick weeknight dinner idea"}'Tracing
When LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY are set, chat emits Langfuse traces (for example recipe-chatbot-session). Local JSON traces also land under the app's data/traces.json.
MCP
MCP evaluate_conversation on the chatbot is not implemented. Do not point the SDK agent at {chatbot}/mcp expecting a real score path — use HTTP chat + Langfuse/paste instead, or the mock MCP for agent demos.