Tonefold lets you edit AI music note by note
A solo developer just open-sourced Tonefold, a free desktop composer for macOS and Windows that turns text prompts into multi-track music. Instead of generating flat audio files, it writes 11 instrument layers directly into an editable piano roll using Claude or local Ollama models. Why it matters: musicians get full control over every note, velocity, and MIDI export without dealing with black-box audio generators.
Tonefold is a free and open-source desktop music maker that turns text prompts into multi-track arrangements you can actually edit note by note. Instead of dumping a finished, uneditable audio waveform onto your desktop like most generative music apps, Tonefold writes raw notes directly into a visual piano roll.
The tool is designed around a simple, practical concept: let LLMs handle music theory and composition logic while keeping the output entirely modular. Describe the vibe, tempo, or chords you want, and it builds out chords, lead melodies, basslines, and drums. It supports 11 distinct layer types in total, adding pads, arpeggios, plucks, counter-melodies, and percussion to the mix, each with its own velocity and timing controls.
What it can do
The biggest selling point is granular control. If you hate a generated bassline, you don't have to re-roll the whole track. You can lock in your favorite chord progression, ask the AI to rewrite only the bass, and keep moving. Whatever layers already exist shape the style and harmonic choices of whatever gets composed next.
Tonefold comes with a built-in General MIDI synth using the GeneralUser GS sound library, meaning you can play and evaluate your tracks instantly without loading external VSTs or opening a complex DAW. When you are happy with the arrangement, you can export plain MIDI files, a stereo WAV mix, or individual stem files layer by layer.
It also includes a Producer mode where an agent proposes a structured roadmap detailing the key, tempo, structure, and the exact layer and section it plans to touch. The app enforces approval at the system level, so no notes land on your roll until you click yes.
How to get it running
The core application and audio engine are written natively in Rust, keeping local playback and editing fast and lightweight. The composition agent runs as a small Node.js sidecar using the Claude Agent SDK, communicating with the desktop app over a local WebSocket.
For intelligence, you can plug in your own Claude account via Claude Code or point the app to a local or networked Ollama instance. If you run Ollama, the entire creative workflow runs completely offline with zero API costs. For DAW users, optional CLAP and VST3 plugins are included, tested primarily in FL Studio.
The project is fully open-source under GPL-3.0, and the license applies to the app code, not the music you export. Universal builds are available for macOS 11 and later covering Apple Silicon and Intel, alongside Windows 10 and 11 installers. Note that the macOS build is not yet notarized by Apple, so you will need to approve it manually under Privacy and Security settings on first launch.
Why it matters
Most generative audio models are black boxes: they spit out complete songs that sound impressive until a single note is out of tune, leaving you with no option other than rolling the dice again. Tonefold flips this dynamic by treating the AI as an interactive MIDI collaborator rather than an audio synthesizer.
We think this is the right direction for creative production tools. By keeping the AI inside the symbolic domain of notes and velocities, it respects the musician's agency and lets creators bring rough AI drafts into their professional DAW pipelines. It is a fantastic free tool for bedroom producers, game developers needing background stems, or anyone frustrated by the limitations of flat audio generators.
Sources (1)
- tonefold.vkfolio.com(opens in a new tab)Originaltonefold.vkfolio.com