How it works
1
Press the shortcut
Meta+Alt+D by default. A small dark pill appears at the bottom of your active monitor.
2
Speak
The pill shows a live level meter and a live transcript, so you can see it is hearing you.
Text appears as you talk, not only at the end.
3
Press it again
Sotto finishes the last utterance, formats everything, and puts the result into the focused
application.
What makes it different
- Local. whisper.cpp runs on your GPU or CPU. Audio never leaves the machine, and there is nothing to sign in to.
- Proper writing. Punctuation and capitalisation come from the model; Sotto adds paragraph breaks when you pause, understands spoken “new line” and “new paragraph”, and strips the noise markers a transcriber emits.
- It goes where you are typing. Text is pasted into the focused app, typed key by key, or left on the clipboard — your choice.
- It never steals focus. The dictation pill is a layer-shell surface: it cannot be focused, and tiling compositors will not tile it.
Requirements
Sotto is early v1, developed and tested on Arch Linux with Plasma 6 on Wayland and an AMD GPU with
ROCm. Expect rough edges elsewhere.
There is no Windows or macOS build. The audio capture, the global shortcut, the text injection and
the overlay are all Linux-specific, so a port is real work rather than a rebuild.
Where to go next
Installing
AppImage, or build with the GPU backend that suits your hardware.
Dictating
The shortcut, the overlay, and getting text into other applications.
Speech models
Which model to download, and what it costs you in speed and disk.
Formatting
Paragraph breaks, voice commands, and what gets cleaned up.
Notepad and CLI
Dictate without inserting, and script Sotto from anywhere.
Privacy and licence
What the local-only guarantee covers, and the licence.