StableDAW Is Now theDAW: What Changed and How to Move Your Setup

This is part two of our local AI music studio series. In part one we covered StableDAW, the browser-based AI music studio built on Stable Audio 3 that installed in one click through Pinokio and ran the small model on a plain CPU. Nine days after that article went up, the project was archived. It did not die, it moved and grew: the repo is now gantasmo/theDAW, it went from seven workspaces to ten, and it picked up VST3 hosting, stem separation, and MIDI transcription along the way.

There is a catch, and it is the reason this article exists. The one-click install path still works, and that is the problem. It silently gives you the old build.

TL;DR

• StableDAW is archived and superseded by theDAW. Same author, same MIT license, roughly twice the application.
• The Pinokio launcher has not been updated. It still clones the archived repo, so the install succeeds and hands you the June build. There is no Pinokio path to theDAW right now.
• Latest release is v0.1.3 (2026-07-06). Use the Windows script, a release installer, the container image, or a source checkout instead.

What happened to StableDAW

The numbers tell the story plainly. The StableDAW repo took its final commit on 2026-06-13 and was archived later that month, sitting at 44 stars. theDAW is a separate, active repo: last push 2026-07-10, 116 stars, and four tagged releases in July. Both are MIT.

This was not a GitHub rename, which matters more than it sounds. A rename leaves a redirect, so old clone URLs quietly follow you to the new project. Here the old repo still exists as a read-only archive at its original address. Anything pointing at it keeps working and keeps serving June code.

Not sure which one you are running? The fastest check is the tab bar. Seven tabs ending in TRAIN means the archived build. Ten tabs including Foundry and Audimate means theDAW. Or ask git directly:

cd /path/to/your/install
git remote -v
# .../gantasmo/stabledaw  -> archived June build
# .../gantasmo/theDAW     -> current

The Pinokio problem

Part one told you to install through Pinokio, because at the time that was the best path available. That advice is now stale, and it fails in the most annoying way possible: quietly.

The launcher at cocktailpeanut/stabledaw.pinokio took its last commit on 2026-06-12, the day before StableDAW's final commit and before the archive. It has not been touched since. Its install step is a single hardcoded clone:

git clone https://github.com/gantasmo/stabledaw app

Because the archived repo is still readable, that clone returns 0. Pinokio reports a clean install. You get a working music studio. It is simply the June build, missing Foundry, Audimate, Perform, VST3 hosting, Demucs stems and MIDI transcription, and it will never update.

We checked for a replacement and there is not one yet: cocktailpeanut/thedaw.pinokio and the obvious variants all 404 as of 2026-07-25. So if you want theDAW today, you install it another way. Below are the paths that actually land on the current code.

Installing theDAW

Four options, roughly easiest first. Requirements across all of them: Python 3.10 or newer, Node.js 20.19+ or 22.12+, FFmpeg on your PATH, and for the Medium model or Magenta RealTime 2, an NVIDIA driver at 550 or later. The Small model still runs on CPU, Apple Silicon included.

Windows: the batch script

Double-click theDAW.bat. It inspects the machine, installs whatever is missing after one confirmation, and opens the app in your browser. There is also theDAW-desktop.bat for the Electron shell rather than a browser tab, and install/setup.ps1 if you would rather drive PowerShell yourself.

Release installers

Every GitHub release ships a Windows installer and a macOS disk image, which is new since part one. Grab theDAW-Setup-<version>.exe or theDAW-<version>.dmg from the releases page. Current version is v0.1.3.

Container

There is now an official image, which is the cleanest option if you already run containers and do not want Python and Node on your host:

docker pull ghcr.io/gantasmo/thedaw
# the repo also ships a docker-compose.yml

From source

The one to use if you want to track development. Note the submodules flag, the checkout is incomplete without it:

git clone --recurse-submodules https://github.com/gantasmo/theDAW
cd theDAW

# backend
uv run uvicorn backend.server:app --host 0.0.0.0 --port 8600 --reload

# frontend, in a second terminal
cd frontend && npm run dev

Port 8600 is not negotiable: the frontend proxies /api/* to it and that address is baked into the Vite config. The frontend takes 5173, and the launcher also clears 5187. Same port story as part one, so if you had those free before, you still do.

The four new workspaces

Underfit is TRAIN with a new name and more depth. It fits LoRA adapters to the Stable Audio 3 RF base models across eight adapter types: lora, dora-rows, dora-cols, bora, and an -xs variant of each. You can filter which layers get adapted with --include and --exclude, adjust adapter strength at runtime rather than retraining, and stack multiple adapters additively. Still needs an NVIDIA GPU.

Foundry is the one with no obvious equivalent anywhere else. It is an infinite-canvas designer for plugin interfaces: you lay out the controls, and the finished design exports as a .gan web-plugin that hosts inside the MIX chain right next to VST3 plugins and the built-in effects. Designing your own plugin UI is not something local AI audio tools normally let you do.

Audimate is a node-graph editor for building generation pipelines. Nodes cover library sources, generation (Stable Audio or Magenta), effects, merges and feedback stages, wired together with bezier edges on a pannable canvas. If you have used ComfyUI for images, the shape will be familiar.

Perform is a live clip launcher fed from imported projects, aimed at playing a set rather than arranging one. Together with the DJ tab's new stems, key-lock and automix, the live half of the app got most of this release's attention.

What else came along

Three additions are worth knowing about even though they do not have their own tabs. theDAW ships a CUDA port of Google's Magenta RealTime 2, which the project describes as the first non-Mac port. Demucs handles stem separation, so you can pull a track apart rather than only generating new material. And audio can now be transcribed to MIDI and rendered as notation, which closes the loop between generated audio and something you can actually edit as music.

Moving your existing setup

There is no migration tool, so treat this as a fresh install alongside the old one rather than an upgrade in place. Keep the archived install until you have confirmed the new one works.

Your generated audio is just files, so copy them out of the old install's data directory. LoRA adapters you trained in TRAIN target the same Stable Audio 3 RF base models that Underfit uses, so they should carry over, though verify strength settings after loading rather than assuming they match. The one thing you cannot bring is the LEARN genealogy graph, which is tied to the old install's library database.

Limitations and gotchas

• No Pinokio path. The headline issue. Until someone updates the launcher, the one-click route lands on archived code.
• Early version numbers. v0.1.3, with four releases in two days, and the repo carries a file named ERRORS-AND-FIXES-2026-07-05.md. This moves fast and breaks.
• The GPU line moved. More of the interesting surface now needs NVIDIA: Medium model, Magenta RT2, Underfit training. CPU users still get the Small model and the whole edit, mix and perform side.
• Licensing is unchanged. App code MIT, model weights under the Stability AI Community License. Commercial use is fine under 1 million dollars in annual revenue. Still no vocals: Stable Audio 3 makes instrumental music and sound effects.
• Submodules. A plain git clone gives you a broken checkout. Use --recurse-submodules.

Who should move, and when

If you installed through Pinokio after part one and have been happy, you are on the June build and you are missing a lot. Moving is worth it, but do it as a parallel install.

If you never installed it, skip StableDAW entirely and start at theDAW with a release installer or the container.

If you only ever wanted a local text-to-audio sketchpad on a laptop with no GPU, the archived build genuinely still does that. There is no urgency.

The ten-minute move: run git remote -v in your install directory. If it says stabledaw, you know where you stand, and the container pull above gets you current without touching your existing setup.

Sources and further reading

• Part one: StableDAW, the one-click AI music studio
• gantasmo/theDAW on GitHub (MIT)
• theDAW releases
• gantasmo/StableDAW, archived
• The unmaintained Pinokio launcher
• Stable Audio 3 Medium on Hugging Face

Tested on: not independently tested. Our bench has no NVIDIA GPU, so we did not run theDAW end to end. Everything above comes from the project's own repositories and release artifacts, verified against the GitHub API on the date below: archive status and commit dates on both repos, the release tags and timestamps, the launcher's clone target read from its install.js, and the absence of a theDAW Pinokio launcher. Feature descriptions are the project's own.
Date checked: 2026-07-25