voxtype

voxtype [options] [command]

voxtype is an offline-first voice-to-text tool for Wayland and X11. Hold a hotkey (default ScrollLock), speak, release, and the transcript is typed at the cursor (or copied to the clipboard). Omarchy ships it as the optional dictation stack (*Install > AI > Dictation*).Transcription runs locally. Engines include Whisper (whisper.cpp), Parakeet, Moonshine, SenseVoice, Paraformer, Dolphin, and Omnilingual. Output goes through wtype on Wayland, with dotool / ydotool / clipboard fallbacks. Compositor bindings should call voxtype record start / stop (or toggle) rather than relying on the built-in evdev hotkey.

-c, --config FILE

Config path (default ~/.config/voxtype/config.toml).

Override the transcription model.

Override the engine (whisper, parakeet, moonshine, sensevoice, paraformer, dolphin, omnilingual).

Override the built-in hotkey.

Press once to start recording, again to stop.

Force clipboard, or clipboard plus simulated Ctrl+V.

Increase verbosity, or quiet (errors only).

~/.config/voxtype/config.toml

Hotkey, audio device, engine/model, output mode (type, clipboard, paste), replacements, and post-process command. voxtype configure (where shipped) edits this interactively.

daemon

Run in the foreground (default when no command is given).

Dependency checks, model download, systemd user unit, Waybar snippet, GPU/ONNX switching.

Control recording from compositor keybindings or scripts.

Transcribe an audio file.

Daemon state for bars. --format json, --follow, --extended.

Show the current configuration.

Continuous meeting transcription and export (Markdown, JSON, SRT, VTT).

Needs PipeWire or PulseAudio, a typing backend (wtype recommended on Wayland), and a downloaded model. Built-in evdev hotkeys require membership in the input group; compositor bindings do not. wtype does not work on GNOME/KDE Wayland (dotool/ydotool fallback). GPU and ONNX engines use separate binaries/packages.

wtype(1), ydotool(1), wl-copy(1)

Source code · Homepage · Documentation