AI Bridge for Ableton Live 1.0.0

An AI that can actually use Ableton Live — not just talk about it.

AI Bridge installs into Live as a Control Surface and lets an AI assistant work your set: load the instrument, set the routing, find the sound you described, write the clip, move the faders, render the stems, tell you how loud it came out. You ask in your own words, and it happens in Live, in front of you.

New to music, or new to Live? It will do the thing for you — or show you where it lives and how it works, which is the part a manual never quite manages.

Been producing for years? Hand it the jobs you have done a thousand times — naming, tidying, routing, comping, stems, loudness checks, hunting for that one sample — and keep the part you actually enjoy.

It knows theory too, so it can help you write when you want that. Useful, but not the point: the point is that it can operate the instrument you already own.

Nothing leaves your computer, nothing happens behind your back, and Ctrl+Z still works.

Install — one line on Windows

irm https://raw.githubusercontent.com/AxidentDK/ai-bridge-for-ableton-live/main/install.ps1 | iex

It downloads the bridge, installs it into Live, and puts Gemini Studio on your desktop. No admin rights — everything lands under your own user profile. It checks for Python 3.10+ and will not install one behind your back.

Prefer to read what you run? Clone the repo and python install.py. Same result.

Then, in Live: Preferences → Link, Tempo & MIDI → Control Surface, pick AI Bridge in an empty slot, Input and Output on None.

Two ways in — pick your window

	Claude	Gemini
Window	the Claude desktop app, or Claude Code	Gemini Studio, included
Connects by	MCP	function calling
You need	a Claude subscription	a Gemini API key — free to start

Gemini reaches the bridge differently from Claude, so it ships with a window of its own. Paste a key once — Settings → Gemini API key, which links to Google's page for a free one — and it can work your set. Tool calls appear as they happen, so a long pause has a visible cause.

⚠️ Gemini is limited on the free tier. It is enough to work a track at a time, but not to hold a whole project in view or to A/B two versions — which is most of what mixing and mastering is. A paid tier is recommended; current limits are in AI Studio.

What it can do

62 tools, and under them a generic proxy reaching 563 operations across 21 object types — every one listed by name in docs/LOM_REFERENCE.md, read from a running Live rather than copied from a manual.

Write and edit clips, play the set, mix, load Live's instruments and effects and turn every knob on them, snapshot a device and morph between two of them, comp takes, render the arrangement or every track as aligned stems, measure loudness in LUFS, and subscribe to changes so the AI is told when something moves instead of asking.

Searching by ear. With the optional listening module, you can ask for "a crisp clap" or "a dark pad" and get sounds chosen by what they sound like. Without it, search falls back to filenames plus Live's own audio similarity — and the answer tells you which was used, so you always know whether something actually listened.

Zero dependencies. Python and nothing else.

What 1.0 means

The tools and the wire protocol are stable from here — anything that breaks them takes a major version. It is not a claim that nothing more will be added.

Known, and written down

• Tested on Windows with Live 12.4. macOS should work and has not been verified — if you run it there, say what happened.
• The Max for Live audio tap is not shipped as a built device (Max bakes an absolute path into it), so three of the 62 tools need Max to build it. Rendering the set and stems does not.
• Tempo estimation is right on 69% of a 546-file ground truth, with 9.7% out by an octave, concentrated at very fast and very slow material. Three approaches to fixing it have been measured and ruled out; the working notes record what was tried and why it failed rather than pretending otherwise.

168 tests green, every tool verified against a running Ableton Live 12.4.3.

v0.5.0-beta — first public beta

First public beta. Everything shipped here is live-validated against a real Ableton Live 12.4.3 on Windows:

• Full LOM reach via five generic primitives — get / set / call / resolve / children (no hand-written method per feature)
• Observers with guaranteed teardown (unobserve, client disconnect, shutdown) — Live's main thread never blocks on a socket
• Clip notes — read / add / remove MIDI notes
• Browser loading — load Live's instruments and effects onto tracks
• Save & export — save the set, render the arrangement to WAV with the rendered duration verified
• Loudness analysis — integrated LUFS (ITU-R BS.1770-4), sample peak, band energies
• Housekeeping — clean up unused tracks, switch main views
• MCP stdio server — zero-dependency, for Claude Code / Claude Desktop and any MCP client
• 43 offline unit tests, no test dependencies

Install: python install.py, then enable AI Bridge as a Control Surface in Live's preferences. See the README for the musician-friendly tour.

Known limits: third-party plugin internals are unreachable (the honest half-open story in the README); the first save of a brand-new set is manual by design.