hellohal2064/cca-notetaker repository overview

⁠CCA Note-Taker

Serves Qwen3.5-9B-FP8⁠ with thinking + tool calling for CCA's note-taker and tool orchestrator roles. Part of the CCA Full Stack⁠.

⁠What This Does

A small, fast LLM that runs alongside the main CCA coder model:

• Note-taker — observes every CCA interaction, extracts insights, stores to Qdrant
• Tool orchestrator — drives multi-step research tool iterations (search, graph queries)
• Summarization — compresses conversation trajectories into structured notes

Runs on the same GPU as CCA (uses only 50% GPU memory), leaving room for the embedding server.

⁠Quick Start

docker run -d \
  --gpus all \
  --name vllm-notetaker \
  -v /data/models:/models:ro \
  -p 8400:8400 \
  hellohal2064/cca-notetaker:latest

⁠Model Required

Download Qwen3.5-9B-FP8⁠ (~9GB) to your models directory.

⁠API

# Chat completion with tool calling (OpenAI-compatible)
curl http://localhost:8400/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "/models/Qwen3.5-9B-FP8", "messages": [{"role": "user", "content": "Hello"}]}'

# Health check
curl http://localhost:8400/health

⁠Configuration (baked into image)

⁠Part of CCA Stack

This is one of 11 containers in the CCA full stack. See:

• CCA agent⁠
• Full deployment guide⁠