Skip to content

beta-todo.md

Latest commit

 

History

History
22 lines (19 loc) · 1.87 KB

beta-todo.md

File metadata and controls

22 lines (19 loc) · 1.87 KB

HiveMind Local Chat Beta Checklist

Progress tracker for getting the HiveMind chat running locally with EMUs per the plan and README.

Completed

  • Reviewed HiveMindplan.md and README.md for architecture and setup expectations.
  • Streamlined chat UI with EMU readiness cards, slash-command helper, and local testing prompts.

To-Do

  • Install prerequisites: Node.js (18+), npm, and Ollama with GPU/CPU support as available.
  • Pull the local router SLM: ollama pull qwen2.5:1.5b-instruct (or Phi-3.5 equivalent) to enable intent routing.
  • Install project dependencies: npm install (bootstraps backend and Vite frontend packages).
  • Configure environment (e.g., .env or config) with any API keys for cloud LLM fallback and ports (server: 4000, web: 5173 by default).
  • Start the backend router/dev server: npm run dev:server and verify http://localhost:4000 responds.
  • Start the frontend chat UI: npm run dev:web and confirm the UI connects to the backend router endpoints (/api/route, /api/chat).
  • Prepare or download sample EMU folder (e.g., poetry.emu/ containing vectors.lance, metadata.json, config.yaml).
  • Implement or validate EMU mount/unmount/list commands (e.g., /mount <emu>, /unmount <emu>, /emus) in the UI/backend.
  • Test retrieval flow with mounted EMUs: router intent → LanceDB hybrid search across mounted EMUs → graded context → cloud synthesize.
  • Run /learn <emu> <text|file> to ingest new content and ensure metadata.json/learn.log update appropriately.
  • Exercise /bench <emu> to record relevance/latency scores and store them in EMU metadata for selection.
  • Log latency/token telemetry during chat sessions to verify performance targets (40–50 tokens/sec on quantized SLMs).
  • Document findings and issues during testing for iterative improvements.