Linux
Experimental, source-only Linux support. There is no supported release.
Linux support is experimental and source-only. There's no supported Linux desktop release.
Warning
Experimental and source-only. The first target (Arch or Omarchy on x86_64, with Hyprland and PipeWire) has had component checks, but physical acceptance is still open. Container and headless checks don't establish microphone or Hyprland support.
What exists
The source includes a daemon and CLI aimed at Linux x86_64 on Arch or Omarchy, with Hyprland and PipeWire.
| Feature | State |
|---|---|
| Delivery | Recovery-only by default, with no clipboard or key changes. Automatic typing and paste are explicit opt-ins. |
| Cleanup | Off by default. Basic is deterministic. Fast and Thorough run local models on the CPU after an explicit prepare step. Thorough is Qwen3 4B here, not the Mac model. |
| Spellings and phrases | Supported. |
| File transcription | One WAV, AIFF or M4A at a time, with your distro's ffmpeg. |
| Live word previews | Not available. |
| Secure-field detection | Not available. |
| Insertion verification | Not available. |
Where to start
The source archive is at mouthtype.app/source. It includes the guides below, which are also copied into these docs.
Linux desktop setup and testing
tools/linux-desktop/README.md.
Omarchy plugin
tools/omarchy-plugin/README.md. It adds a bar menu, a floating pill and a choice of toggle or hold shortcut. Its installer enables a login service unless you pass --no-autostart.
Agent install checklist
an ordered install checklist for people and coding agents.
The speech-preparation helper downloads pinned model files, verifies their hashes and builds a dedicated environment. It doesn't record, register shortcuts or start a service.