AI MIDI generation for Ableton Live

Download on maxforlive.com · Support this project on Ko-fi

Conduit is a Max for Live device that connects Ableton Live to a local AI running on your own machine. Type a prompt like "generate a 4x4 techno drum beat" and it writes MIDI clips directly into your session. No cloud API keys, no subscriptions, no sending your music anywhere. Everything runs locally and privately through Ollama.

Conduit has two modes:

Talk to the AI about your session. Ask for ideas, get feedback, discuss arrangements. It can see your BPM, time signature, track names, and more — so the answers are relevant to what you're actually working on.

Tell the AI what you want and get MIDI clips written straight into your selected track. Drums, melodies, basslines, chord progressions — whatever you need to get moving.

Genre-aware — 8 built-in genre modules (techno, house, dnb, dubstep, hip-hop, ambient, IDM, trance) that shape how the AI thinks about rhythm, velocity, and note choice. Pick one from the dropdown.

Session-aware — The device polls Ableton every 5 seconds for BPM, time signature, track names, selected track, and transport state. The AI uses all of this as context when generating.

Pattern clipboard — Every pattern you generate is auto-saved. Hit Paste to re-insert the last one into a new clip slot. Up to 20 patterns stored per session.

Undo — Changed your mind? Revert the last generated clip instantly.

100% local — Runs on Ollama with llama3.2 (3B parameter model). No API keys. No cloud. Works offline once you've pulled the model.

• macOS (Apple Silicon or Intel) or Windows
• Ableton Live 11 or 12 with Max for Live
• Python 3.9+
• Ollama (free, open source)
• ~4GB free RAM for the model

Download from ollama.com, drag it to Applications, and open it once so it finishes installing.

Open Terminal and run:

ollama pull llama3.2

This downloads the AI model (~2GB). Only needs to happen once.

Open Terminal in the project folder and run:

./package-device.sh --install

This builds the device and copies it to Ableton's User Library.

Double-click Start Conduit.command. It installs dependencies, starts the server, and keeps it running. Keep this window open.

Open Ableton's Browser, go to User Library > MIDI Effects > Conduit, and drag it onto a MIDI track. Wait a few seconds for the device to connect, then type a prompt and press Enter.

Download the Windows installer from ollama.com/download and run it. After installation, Ollama runs in the background automatically.

Download Python 3.9+ from python.org/downloads. During installation, check the box that says "Add Python to PATH" — this is important.

Go to github.com/jordanaftermidnight/conduit, click the green Code button, then Download ZIP. Extract the ZIP somewhere easy to find, like your Desktop or Documents folder.

You need to copy two files from the Conduit folder into Ableton's device folders. Open File Explorer and do the following:

Copy the device file:

1. In the Conduit folder, go to dist > Conduit
2. Copy Conduit.amxd
3. Paste it into: C:\Users\YourName\Documents\Ableton\User Library\Presets\MIDI Effects\Max MIDI Effect\Conduit\

Copy the bridge script:

1. In the Conduit folder, go to m4l
2. Copy conduit-bridge.js
3. Paste it into both of these locations:
   • C:\Users\YourName\Documents\Ableton\User Library\Presets\MIDI Effects\Max MIDI Effect\Conduit\
   • C:\Users\YourName\Documents\Max 8\Packages\Conduit\javascript\

Note: Replace YourName with your actual Windows username. If any of these folders don't exist yet, create them yourself — right-click > New > Folder.

In the Conduit folder, double-click Start Conduit.bat. The first time you run it, it will download the AI model (~2GB) and install Python dependencies — this only happens once. After that, you'll see a message saying the server is running. Keep this window open while you use Conduit.

Open Ableton Live, go to Browser > User Library > MIDI Effects > Conduit, and drag the device onto a MIDI track. Wait a few seconds for it to connect (the status bar will say "Connected"), then type a prompt and press Enter.

Every time you want to use Conduit, double-click Start Conduit.bat before opening Ableton.

┌─────────────────────────────────────────────┐
│  CONDUIT       [chat/generate ▾]  [genre ▾] │  <- mode + genre dropdowns
│                                              │
│  [ type your prompt here...              ]   │  <- prompt input
│                                              │
│  ┌────────────────────────────────────┐      │
│  │  response display area             │      │  <- AI response
│  │                                    │      │
│  └────────────────────────────────────┘      │
│                                              │
│  [Reset] [Generate] [Paste] [Undo] [Clear]   │  <- buttons
│  ● Connected                                 │  <- status bar
└─────────────────────────────────────────────┘

1. Select a MIDI track in Ableton (or create a new one)
2. Set mode to generate in the dropdown
3. Pick a genre
4. Type what you want — "4-bar techno kick pattern", "8-note ambient melody in C minor", whatever comes to mind
5. Press Enter or click Generate
6. The clip appears in the first empty slot on your selected track

1. Set mode to chat
2. Ask anything — "what scale would work over these chords?", "suggest a drum fill for the bridge", "how should I arrange this into a full track?"
3. The AI sees your session context so it can give answers that actually make sense for your project

The more specific you are, the better the results.

Be specific about what you want:

"16-note hi-hat pattern with offbeat accents" beats "make some hats" every time.

Mention note counts and length:

"8 notes", "16 steps", "4-bar phrase", "2 bars of..."

Reference genre conventions:

"303-style acid bassline", "amen break variation", "four-on-the-floor with ghost snares"

Use music theory if you know it:

The AI understands scales, intervals, chord names, and rhythmic subdivisions. Don't be afraid to get technical.

Keep it conversational if you don't:

"something dark and minimal" or "funky bassline that grooves" works too. The genre module fills in the rest.

conduit/
├── server/                  # Python server (FastAPI + Ollama)
│   └── genres/              # Genre module YAML files
├── m4l/                     # Max for Live device source
├── docs/                    # Technical documentation
├── tests/                   # Test suite
├── assets/                  # Logo and branding
├── Start Conduit.command    # macOS launcher (double-click)
├── Start Conduit.bat        # Windows launcher (double-click)
└── package-device.sh        # Build + install script (macOS)

For the full technical breakdown — architecture, API reference, AMPF format spec, signal flow, and how to extend Conduit with custom providers and genres — see docs/ARCHITECTURE.md.

"Server not responding" Make sure the server is running — Start Conduit.command on macOS or Start Conduit.bat on Windows. Check the terminal window for errors.

"No model found" You need to pull the model first. Open Terminal and run ollama pull llama3.2.

Device shows "initializing..." Give it about 5 seconds. The device needs a moment to handshake with the server.

No MIDI output Make sure you have a MIDI track selected in Ableton and the device is set to generate mode, not chat.

Clips sound wrong or unexpected Try being more specific in your prompt. Mention the key, scale, number of notes, and rhythmic feel you're going for. Switching genres can also make a big difference.

MIT License. Copyright (c) 2026 Jordanaftermidnight.

Free to use, modify, and distribute. See LICENSE for details.

Conduit v1.0 — your session, your AI, your machine. 🎹