Download
Free, open source, and yours to keep. Runs entirely on your computer.
All downloads are served from GitHub Releases, with checksums published alongside every build. A free Steam release is in preparation and will carry the same local-first guarantee.
What to expect on first launch
No model ships in the box — you choose what runs on your machine. On first launch the app walks you through downloading a starter model (Qwen3 4B Instruct, ~2.5 GB, Apache-2.0), with the license and size shown before anything is fetched. One model download is the only time the app needs the internet. After that, everything works offline.
Recommended hardware: any 64-bit machine with 8 GB RAM runs the starter model; 6 GB+ of VRAM unlocks the larger, sharper models. Full details in the install guide and local models guide.
For developers
The whole platform is open source (Apache-2.0) on GitHub — see the developer install guide to run it from source and the contributing guide to get involved.
Having trouble?
The troubleshooting guide covers the common failure modes, and GitHub issues are open to everyone. Beta builds include a one-click, fully-redacted diagnostic report — nothing is ever uploaded automatically.