M6: Enhanced Health Check with LLM and RAG Status

[kovtcharov]

Priority: P2 — Operational reliability

Effort: Small (1-2 days)

Problem

/health returns {"status": "ok"} when the FastAPI process is running, even if Lemonade Server is down or no model is loaded. Webapps that check health before accepting traffic get false positives.

Deliverable

Extended health endpoint that reports component status:

GET /health
{
  "status": "ok",
  "components": {
    "api": { "status": "ready" },
    "llm": {
      "status": "ready",
      "backend": "lemonade",
      "model": "Qwen3.5-35B-A3B-GGUF",
      "url": "http://localhost:8000"
    },
    "rag": {
      "status": "ready",
      "documents": 3,
      "total_chunks": 247
    }
  }
}

Status values: ready, unavailable, error, not_configured

Files to change

• src/gaia/api/openai_server.py — extend /health handler
• Ping Lemonade at $LEMONADE_BASE_URL/v1/models with a short timeout
• Report RAG status if M2 (M2: Document Indexing API on OpenAI-Compatible Server #651) is implemented

Acceptance criteria

• /health returns component-level status
• Returns "status": "degraded" if LLM is unavailable but API is up
• Response time under 500ms (short timeout on Lemonade ping)
• Backward compatible — top-level "status" field still works for simple checks

Context

Full milestone plan: docs/plans/webapp-integration.md