What is ableton-mind?
ableton-mind lets an AI assistant build and inspect an Ableton Live set from plain language. You describe the musical result; the assistant uses real Live tools, reads the set back, and reports what changed.
"Create a 128 BPM tech-house starter: four-on-the-floor kick, rolling bass, off-beat hats, a sidechain rack, and a 16-bar drop."
The goal is not a chat-only suggestion. The goal is a real Live set: tracks, clips, devices, automation, mixer state and a verification diff.
Who it is for
- Producers who want a fast starting point inside Live instead of a blank session.
- Live performers who want repeatable rigs, clip-launching layouts and Push/Launchpad-friendly setups.
- Sound designers who want the assistant to use real device names and parameter ranges.
- Developers who want a typed MCP bridge into the Live Object Model.
If you want the internals, jump to the developer architecture. If you want to make music first, start with Your first Live set.
Why it works
Most AI music workflows stop at advice. ableton-mind is designed around three grounded layers:
- Real execution: a Python Remote Script runs inside Live and talks to the TypeScript MCP server over local JSON-RPC.
- Real knowledge: 55 Live device schemas, scales and recipe metadata are embedded so the assistant can use valid parameters instead of guessing.
- Real verification: tools read state before and after mutations and return
{ ok, verified, diff }.
Current status
The code has passed a real Ableton Live 12.4.1 smoke test for core transport/session/track operations. Public npm and .mcpb release channels are still blocked until TD-048 package validation is green, so use the source install path for now.
Next steps
- Install from source.
- Make your first Live set.
- Keep the prompt cookbook open for starting prompts.