Ableton Composer

Generate structured MIDI for Ableton Live with multiple AI providers, reusable style profiles, and modular planning prompts.

System overview

Prompt, analyze, compare, push.

Ableton Composer is built like a toolchain, not a toy demo. Generate MIDI structures, derive style bundles from source material, and drive the next generation pass with tighter harmonic and arrangement context.

Start here See workflows
Providers Anthropic · OpenAI · Codex · Claude CLI
Profiles song · album · artist · collection
Planning harmonic · arrangement · blueprint
Runtime Ableton Live via ableton-js
Generation
Prompt to song JSON
Create multi-section sets from natural language, track lists, and style bundles.
Analysis
Song to profile bundle
Extract album, artist, and collection fingerprints across harmony, rhythm, and arrangement.
Runtime
Push into Live
Write clips directly into Ableton through the `ableton-js` remote script workflow.

Quick Start

git clone https://github.com/delaudio/ableton-composer
cd ableton-composer
npm install
cp .env.example .env

Generate a first set:

ableton-composer generate "melancholic IDM, 8 sections" \
  --tracks "Bass,Drums,Pad,Lead,Chords,FX" \
  --provider openai

Start Here

Entry
Getting Started
Install the CLI, set provider keys, and run the first generation.
Reference
CLI
Command surface for generation, analysis, compare, expand, and presets.
System
Profiles
Understand bundles, prompt profiles, and hierarchical style scopes.
System
Prompt System
See how overlays and planning stages shape final generation.
Practice
Workflows
Use end-to-end flows for albums, presets, expansions, and compare loops.
Ops
Deploying
Build and ship the docs site through Minuto and Vercel.

Recommended Paths

I want to generate material quickly

  1. Open Getting Started
  2. Use generate with explicit tracks
  3. Push the result into Live

I want to adapt a reference bundle

  1. Read Profiles
  2. Run analyze --scope album
  3. Generate from bundle.json
  4. Validate with compare

I want to understand the prompting system

  1. Read Prompt System
  2. Inspect genre overlays and harmony overlays
  3. Review the planning layers: harmonic, arrangement, blueprint

Core Capabilities

  • generate creates song JSON from prompts, style bundles, and planning prompts
  • analyze extracts hierarchical style profiles from songs, albums, artists, or collections
  • compare measures alignment and drift between generated material and a reference profile bundle
  • expand adds new tracks to existing sections without rewriting the originals
  • preset generate creates synth presets from preset profiles

Architecture Snapshot

  • multi-provider AI layer for Anthropic, OpenAI, Codex, and Claude CLI
  • modular prompt system with genre overlays, harmony overlays, and planning stages
  • prompt-ready profile bundles for large album-scale conditioning
  • chunked generation for larger structured outputs

Example Workflow

# 1. Analyze your own reference set into a style bundle
ableton-composer analyze sets/reference-collection \
  --scope album \
  --artist "Example Artist" \
  --album "Midnight Signals"

# 2. Generate from the reference bundle
ableton-composer generate "moody electronic sketch with restrained hooks" \
  --style profiles/albums/example-artist/midnight-signals/bundle.json \
  --tracks "Bass,Drums,Pad,Lead,Chords,FX" \
  --provider openai \
  --sections 4 \
  --chunk-size 2

# 3. Compare the output against the reference profile
ableton-composer compare \
  profiles/albums/example-artist/midnight-signals/bundle.json \
  sets/my-generated-song

Use reference sets only when you own the material, have permission to analyze it, or are working with synthetic/example material.

Edit this page