Local copilot

ableton-mind includes a local LLM copilot for simple Ableton Live tasks without a paid API. It talks to any OpenAI-compatible chat endpoint and defaults to local Ollama.

Quick start

# one-time: install Ollama, then optionally pre-pull the default model
ollama pull qwen2.5:3b

# from a built checkout
npm run build
node dist/index.js chat

The chat UI opens at http://127.0.0.1:4142.

Safety tiers

The default tier is safe, which is read-only. This is intentional: Live sessions can contain real work, so the local model should inspect before it is allowed to change anything.

ableton-mind chat              # safe: inspect only
ableton-mind chat --write      # standard: simple transport/track/clip/device changes
ableton-mind chat --creative   # standard + recipes/browser load
ableton-mind chat --read-only  # lock the browser UI to safe mode

Headless prompts

Use ask for scripts or quick checks:

ableton-mind ask "List the tracks and tell me what looks missing"
ableton-mind ask --write "Set the tempo to 128 BPM"
ableton-mind ask --json "What can you inspect without changing Live?"

chat --prompt uses the same engine:

ableton-mind chat --prompt "What is in this set?"

Configuration

Var	Default	Purpose
ABLETON_MIND_LLM_BASE_URL	http://127.0.0.1:11434/v1	OpenAI-compatible endpoint
ABLETON_MIND_LLM_MODEL	qwen2.5:3b	Model id
ABLETON_MIND_LLM_API_KEY	unset	Optional bearer token
ABLETON_MIND_LLM_TIER	safe	safe, standard, or creative
ABLETON_MIND_LLM_MAX_STEPS	8	Max model/tool loop steps
ABLETON_MIND_LLM_TEMPERATURE	0.4	Sampling temperature
ABLETON_MIND_CHAT_PORT	4142	Local browser UI port

Flags --model, --base-url, --port, --no-ollama, and --no-open override the defaults for one run.

Bridge behavior

The local UI can start even when Ableton Live is closed. Static guidance still works, but tools that need Live return a bridge-offline error until the AbletonMind Remote Script is active.

The browser server binds to loopback only and rejects non-loopback Host/Origin requests.