Orchestration State 1
Related plan: Orchestration Plan 1
- ✅Task 1 — Enable and Configure the Agents API
- ✅Read current librechat.yaml
- ✅Add remoteAgents interface config
- ✅Redeploy LibreChat
- ✅Generate API key and store as AGENT_API_KEY in Railway env vars
- ✅Verify endpoint is reachable via curl
- ✅Record full response and document agent ID format
- Agent ID format:
agent_<alphanumeric>(custom string, NOT MongoDB ObjectId) - 11 agents returned, e.g.
agent_69d69e3198cb9ee870cd0373(Forge GLM 5.1 [Master])
- Agent ID format:
- ✅Task 2 — Map the Full Chat Completions API Surface
- ✅Send minimal Chat Completions request
- ✅Record full response shape (conversation ID, usage, finish_reason, metadata)
- ✅Test streaming mode, capture SSE event types and data shapes
- ✅Test additional parameters (temperature, max_tokens, response_format, stop, top_p)
- ✅Check LibreChat source code for Chat Completions route handler
- ✅Task 3 — Map the Full Responses API Surface
- ✅Read Open Responses specification
- ✅Send minimal Responses request
- ✅Record full response shape and compare with Chat Completions
- ✅Test streaming mode and capture SSE event types
- ✅Test additional parameters from Open Responses spec
- ✅Check LibreChat source code for Responses route handler
- ✅Task 4 — Investigate Conversation Lifecycle and Tracking
- ✅Check if response includes conversation ID
- ✅If no ID in response, try discovery methods (MongoDB, SSE headers, response headers)
- ✅Test conversation querying/history/follow-up/title/visibility
- ✅Test conversation continuity
- ✅Task 5 — Test Long-Running Agent Invocations
- ✅Select/create test agent with multiple tool calls
- ✅Run with stream: true and measure execution time, SSE behavior
- ✅Run with stream: false and measure HTTP response behavior
- ✅Test disconnect resilience
- ✅Test deliberately long task (15+ min)
- ✅Document all findings
- ✅Task 6 — Investigate the LibreChat UI Chat API
- ✅Reproduce the UI API call and confirm conversationId/streamId in response
- ✅Verify message persistence (user + assistant + tool calls) in MongoDB
- ✅Verify transcript visibility in Chat History
- ✅Test streaming behavior (real-time vs burst)
- ✅Test conversation continuation with conversationId/parentMessageId
- ✅Test long-running execution with full transcript preservation
- ✅Investigate auth mechanism (JWT expiry, refresh flow, API key alternative)
- ✅Check LibreChat source code for UI chat route handler
- ✅Task 7 — Compile Findings and Update Research
- ✅Compile findings from Tasks 1–6
- ✅Update Orchestration_Research.md sections with evidence-based verdicts
- ✅Add “Agents API Operational Profile” section
- ✅Add “UI Chat API vs OpenAI-Compatible API” section
- ✅Propose revised architecture if findings change recommendations
- ✅Update all pending decisions