subtopic

Async calls, streaming, timeouts

Add POST /v1/chat that calls your LLM function. Use async so one slow LLM call does not block other users. Add a streaming version that sends the answer word by word (Server-Sent Events), set timeouts, and return clear errors (503 if the LLM provider is down, 504 on timeout). Return sources with the answer.

Builds on: 3.6 Run it reliably