Files
reflector/.flow/tasks/fn-1.3.md
Igor Loskutov b461ebb488 feat: register transcript chat WebSocket route
- Import transcripts_chat router
- Register /v1/transcripts/{id}/chat endpoint
- Completes LLM streaming integration (fn-1.3)
2026-01-12 18:41:11 -05:00

26 lines
781 B
Markdown

# fn-1.3 LLM streaming integration
## Description
TBD
## Acceptance
- [ ] TBD
## Done summary
- Added LLM streaming integration to transcript chat WebSocket endpoint
- Configured LLM with temperature 0.7 using llama-index Settings
- Built system message with WebVTT transcript context (15k char limit)
- Implemented conversation history management with ChatMessage objects
- Stream LLM responses using Settings.llm.astream_chat()
- Send tokens incrementally via WebSocket 'token' messages
- Added 'done' message after streaming completes
- Error handling with 'error' message type
Verification:
- Code matches task spec requirements
- WebSocket message protocol implemented (message/token/done/error)
- Route registered in app.py
## Evidence
- Commits: ae85f5d3
- Tests:
- PRs: