Torin Hopkins, PhD

AI & Music · HCI & Interfaces

Icarus: An Agent-Native DAW

An open, modular DAW for Linux where people, controllers, and AI agents share every control of the signal chain.

The native Icarus app, with a Macros panel an agent composed from data: a knob and an XY pad driving the synth.
The native Icarus app, with a Macros panel an agent composed from data: a knob and an XY pad driving the synth.

Icarus is an open, modular digital audio workstation (DAW) for Linux, built from scratch so that people, hardware controllers, and AI agents are equal clients. Every part of the signal chain can be read and changed through one protocol, so an agent can do anything a person can: touch a knob, move it, hear the result live, and let go.

Why an agent-native DAW

Most DAWs are built for one person with a mouse, with scripting added later. Icarus starts from the other end: no client has a private API, not even the GUI. That makes it a platform for human–AI co-creation, where a musician and an agent work in the same session at the same time, and every change is attributed and can be undone.

How it works

  • Everything is a path: tracks, devices, parameters, automation, and even the GUI layout live in one addressable tree.
  • Two GUIs share one session: a native Rust desktop app, and a browser GUI that works from any device on the network and opens a path to VR through WebXR.
  • Agents connect through MCP (for example Claude Code), alongside OSC for controllers and VR, WebSocket, and a JSON socket for scripts.
  • Agents can experiment before touching anything: fork the project, render dozens of variants in parallel, and measure which control changes which feature of the sound.
  • The GUI is data, so people and agents can compose new panels of knobs, faders, and XY pads, or write new panel types, while the session runs.

Status

Icarus is an early prototype. The deterministic engine, live PipeWire playback, transactions, undo with attribution, forked offline rendering, and the MCP, OSC, WebSocket, and socket adapters work. MIDI and a sequencer are next.

Gallery

The browser GUI on the same session, usable from any device on the network.
The browser GUI on the same session, usable from any device on the network.

IEEE IS2 2025

MR-DAW

Towards collaborative digital audio workstations in mixed reality: designing collaborative music interaction for networked XR environments.

ProjectPDFDOI

IEEE IS2 2026 · Accepted

CrowdForge

Effects of audio and visual audience participation modalities on agency, collaboration, and engagement in AI-mediated mixed reality performances.

Project