An AI-native audio platform, built around musicians.
Not everything made with AI. AI where it helps: hear an idea sooner, build the tool you imagine, and spend more of your time making music.
Our vision
The glue between models, tools and musicians
Music AI moves fast, but it arrives in pieces: a model in a paper, a demo in a notebook, a plugin that does one thing. We connect those pieces with audio and MIDI tools, playable interfaces and real release targets, so developers can build with them and musicians can play them.
The goal is more people making more music, with tools that answer to their ideas.
Models
- Generation
- Separation
- Transcription
- Neural synthesis
- Embeddings
Maru
- Engine & SDK
- Audio & MIDI tools
- Interfaces
- Studio agent
Where you play
- Browser
- Your DAW
- Your device
AI that serves your expression
How we decide where AI belongs in a musical tool.
You stay the author
AI proposes, you decide. Every result is a starting point you choose, shape, and play.
Material, not final tracks
Sounds, kits, MIDI, presets and source code you keep editing, rather than finished songs.
Open about the models
Model cards show what a model does, its licence, what it was trained on, and its limits.
Runs where you work
In the browser, on your own device where it can, or hosted when a model needs more.
What AI is good for
The tasks where AI gives musicians more room
We want these capabilities in the hands of every developer and musician. Each one below links to something you can try today.
Generate sounds
Describe a sound and get one-shots, textures and source material to shape further.
Build cohesive kits
Collections whose sounds belong together while each plays a different role.
Separate stems
Pull a mix apart into aligned stems to remix, practise, or sample from.
Turn audio into MIDI
Play or sing a line, inspect the detected pitches, and keep editing it as notes.
Suggest and vary MIDI
Melodies, harmonies and variations you can steer, accept, or rewrite.
Play neural instruments
Instruments whose timbre comes from a trained model, played like any synth.
Analyse and discover
Track beats, read pitch, and find sounds by what they sound like.
Build the tool you imagine
Describe an instrument or effect and a coding agent writes real, inspectable source.
For developers
Use the latest AI models
New music AI models appear every month, usually as a research demo. Maru adds everything around them: sound in and out, controls, an interface, and a way to release. They become instruments, effects and tools people can open and use. Bring your own model, or start from one of the 32 in our directory.
- 1Your modelA checkpoint, a script, or a paper's reference code.
- 2Audio & MIDI I/OInputs, outputs, playback and parameter controls.
- 3JobsProgress, cancellation, seeded variations and export.
- 4InterfaceControls, meters, piano rolls and scopes from the UI kit.
- 5ReleaseBrowser, plugin and other targets its runtime supports.
Research
We write papers, and ship what we learn
Text-to-Kit, our method and benchmark for generating playable drum kits from text, is accepted at the ISMIR 2026 Late-Breaking/Demo session in Abu Dhabi. At ADC 2026 in Bristol, our virtual poster examines agent-assisted plugin creation.
We share methods, benchmarks and interactive examples so others can examine the work and build on it.
What we are exploring next
- More control over MIDI generation, with small local models
- Musical agents that listen, anticipate, and play alongside musicians
- Sound embeddings for sample discovery and synthesis
More music, made by more people
Start creating with Studio, or bring us a model, a method or an idea you want musicians to play.