Skip to content

AI Music Studio

A multi-agent system for generating MIDI backing tracks.

AI musician and engineer agents collaborate to produce genre-aware songs with drums, bass, guitar, and keys. Plug in an LLM provider for guided generation, or run fully algorithmically.

Why AI Music Studio?

Problem Solution
Writing MIDI parts for every instrument is tedious Agents generate parts automatically, reacting to each other
Algorithmic music often sounds mechanical Genre-aware patterns + LLM guidance for human feel
Exporting to your DAW is a chore Built-in FluidSynth, TiMidity, GarageBand, Logic Pro, and raw MIDI
Hard to iterate on arrangements REST API and CLI for fast experimentation

Feature Overview

Feature Overview

  • 🥁 Drummer — genre-aware kick, snare, and hi-hat patterns
  • 🎸 Bassist — root-note bass lines locked to drums and chord changes
  • 🎸 Guitarist — rhythm/lead guitar parts and power chords
  • 🎹 Keyboardist — chord voicings, pads, and arpeggios
  • 🎚️ Mixer — per-track volume, pan, EQ curves
  • 🔊 Mastering — final loudness and metadata
  • 🤖 LLM support — OpenAI, Anthropic, or Google Gemini for guided generation
  • 🎵 Google Lyria 3 — full-length AI music generation via MusicGenerationAgent
  • 🔌 Multi-provider system — ProviderRegistry routes requests to the best available backend
  • 🔊 Audio analysis & TTS — AudioAnalysisAgent and TTSAgent via the Gemini package
  • 🛠️ MCP Server — expose generation tools for GitHub Copilot, Claude Code, and other AI assistants
  • 🌐 REST API — FastAPI server for programmatic session control
  • 🖥️ Web UI — lightweight browser interface served alongside the API
  • 🎛️ DAW export — FluidSynth, TiMidity, GarageBand, Logic Pro, MIDI/WAV

Quick Example

pip install -e "."
python scripts/generate_demo.py --genre classic_rock --key E --mode minor --tempo 120 -v

Output lands in ./output/ as <session-id>_full.mid plus individual per-instrument files.

Next Steps