Skip to content
user1303836Public

About

A multi-DAW agent that does literally everything

Topics

Resources

Stars

19 stars

Watchers

1 watching

Forks

Latest commit

 

History

1,377 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

kumi

CI Release Ableton Live 12 Native Rust runtime License: MIT

English · 简体中文 · 日本語

A studio producer agent for Ableton Live that learns how you work. Kumi will literally do anything in your set that Ableton exposes via Extensions/LOM... and even some things it doesn't intentionally expose via the API. From the tedious jobs to the ones you never have time for: rebuilding a sound from a YouTube tutorial, matching your mix to a reference, writing a Max for Live device you describe, or reworking the rack you point at. Every change shows up in HISTORY, most with their own undo, and a whole plan is one undo in Live (Cmd-Z, or Ctrl-Z on Windows). Kumi remembers the techniques you keep, so it fits you better with every session. Support for other DAWs is planned.

Demo_1-readme.mp4

Quickstart

Needs Ableton Live 12 on macOS 13+ or Windows 10/11.

macOS (Terminal):

curl -fsSL https://github.com/user1303836/kumi/releases/latest/download/install.sh | sh

Windows (PowerShell):

irm https://github.com/user1303836/kumi/releases/latest/download/install.ps1 | iex

Kumi opens once it's installed and walks you through signing in and connecting to Live. After that, just run kumi. Something off? kumi doctor says what to fix.

Guide · Commands, keys and screens · How Kumi changes your Set · Changelog

What it does

  • Changes almost anything in the Set: tempo, scale and groove; the mixer, routing and sidechains; tracks, scenes and clips; notes and MIDI transforms; devices, racks and their parameters.
  • Listens: loudness, tonal balance, width, tempo and key of a mix, a sample or its own bounce, and how your mix compares with a reference.
  • Judges its own changes: brings each change back to the level it belongs at (with the last Live Limiter, else a Utility it adds), hears before and after, and keeps a change only if its target improved and nothing else got worse. /loop works in rounds until a goal is met; /goal keeps at one across turns.
  • Measures a style: a file, a folder, a YouTube or Spotify link, or words like "Burial" or "dub techno" become example tracks and a measured profile with the style's range, kept for next time.
  • Matches a reference: builds several takes on a sound, scores each against the reference and refines the best. /loop keeps at it until it gets there.
  • Watches tutorials from YouTube or a file, then builds what they show on a new track.
  • Plays, records and resamples, to let you hear what it built or to check its work.
  • Takes full control of Live: deletes what you ask for, writes MIDI straight into the Arrangement, renders offline, runs Python inside Live for what its other tools can't reach, and answers for what you right-click in Live ("Ask Kumi about this"). Big Sets of hundreds of tracks stay quick.
  • Makes Max for Live devices you describe in plain words (MIDI effects, audio effects and instruments) and puts them on your tracks.
  • Looks things up: searches the web and reads pages, PDFs, manuals and code on GitHub, so it can build an effect like one it has read about.
  • Shows where you are: FOCUS follows what you touch in Live, as a device tree, a piano roll or a Session or Arrangement strip. Click a device to point at it: "this Saturator's too harsh".
  • Remembers: notes about you and each Set, techniques it learns from what you keep, and recipes you can replay. Every save is shown, and one click forgets it.
  • Keeps your conversations for each Set, each with an id that takes it up again from anywhere (kumi -r), and says what changed while it was closed. /share turns one into a web page to send, with keys and paths taken out.
  • Works with your model: sign in with ChatGPT, or use an OpenAI, Anthropic or OpenCode API key.

Status

  • Kumi 1.19.3 is currently supported for Ableton Live 12.4.15b4 (beta) and above on macOS and Windows. You may run into issues on lower versions. If you do, please file an issue!
  • 1.19.3 is tested with Live on macOS. On Windows, installing and updating are tested; using it with Live there is still new. Please send a kumi report when something breaks.
  • Support for Reaper and Renoise are planned

Development

Build this checkout with Rust and Cargo:

cargo build --release --locked --workspace --bins
cargo run --release -p kumi --                     # add bridge, doctor, or other arguments after --
npm ci --prefix crates/kumi-runtime/tests/support  # once: the official SDKs some tests run
sh scripts/test-isolated.sh                        # no Live or sign-in needed

Existing npm run setup and npm run kumi -- ... commands remain available. With Cargo installed, these build and run this checkout. Without Cargo, they install and launch the matching published native release, keeping the same settings, sign-ins, conversations and library in ~/.kumi. After that handoff, kumi runs the native application directly.

Folder What's in it
crates/kumi The native terminal app and kumi command
crates/kumi-runtime Model providers, memory, audio analysis, video, the web, and the Live integration
crates/ableton-mcp-server The native bridge, also usable by other MCP clients (bridge guide)
remote-script The bridge's Remote Script, which runs inside Live
apps/live-extension Kumi's Live extension (Live 12.4 and later)
protocol The list of operations the bridge and the Remote Script share

The developer guide covers building, testing and releasing.

License

MIT; the parts from other people's work, and the models Kumi fetches, are in THIRD_PARTY_NOTICES.md. Ableton Live is a trademark of Ableton AG; Kumi is not affiliated with or endorsed by Ableton.

About

A multi-DAW agent that does literally everything

Topics

Resources

Stars

19 stars

Watchers

1 watching

Forks

Releases

Used by

Contributors

Languages