Voice
Set up local dictation with whisper.cpp, dictate into panes and fields, and control PaneScale with spoken commands in English and German — all on your computer.
Voice has two jobs: dictation — speech becomes text where you are working — and commands — short phrases that act on PaneScale itself. Both run on your computer, even offline.
Set up
- Open Settings → Voice & dictation.
- Click Install whisper.cpp. PaneScale downloads the official release only when you click; a progress bar shows the download.
- Download a model: Whisper Base (142 MiB, fast) or Whisper Turbo (547 MiB, more accurate). Both are multilingual. With both downloaded, choose which one is active.
- Pick a language — automatic, English, German, Arabic, French or Spanish.
Models are downloaded over HTTPS and verified against a pinned SHA-256 checksum and size before use. When everything is in place, the settings say Dictation is ready — it runs entirely on this computer, even offline.
The voice panel
Click the microphone in the title bar. The panel has two modes — Dictation and Command — and shows local · Whisper as a reminder that nothing leaves your machine.
Click the large microphone, or press Space, speak, and click again. The status goes from Microphone starting… to Listening to Transcribing locally…. A recording can be up to ten minutes long; it is saved to a temporary file for transcription and deleted right after.
Dictate
The panel tells you where the text will go:
| Target | When |
|---|---|
| The active input field | A text field in PaneScale had focus — the chat composer, a workflow brief, a task description |
| The focused pane | A terminal pane had focus — the text is typed into it, without pressing Enter |
| The clipboard | Nothing could take it — you can paste it anywhere, including other apps |
The microphone in the chat composer starts dictation straight into that conversation.
Commands
Switch the panel to Command and speak. Commands are understood in English and German, filler words such as “please” or “hey” are ignored, and you can chain two with “and” or “then” — “open code and show four panes”.
| Say | What happens |
|---|---|
| “open code” · “chat mode” · “show workflows” | Switches the mode: agent, code, chat, workflow, video |
| “open autopilot” · “open settings” | Opens the stage |
| “open tasks” · “open routines” · “open hub” | Opens the overlay |
| “open browser” · “open dashboard” | Opens it in the side panel |
| “four panes” · “show 6 panes” | Switches to the smallest layout with at least that many panes |
| “start claude” · “start codex” · “new terminal” | Opens a new pane with that agent — Claude, Codex, Gemini, Copilot, Aider — or a shell |
| “what needs me” | Opens the dashboard with everyone who is waiting for you |
| “new task empty state for uploads” | Creates a card on the tasks board |
| “tonight update the readme” · “for tonight …” | Creates a card marked for tonight’s autopilot queue |
| “stop all” · “stop tonight” | Stops running task runs |
| “theme” · “dark” · “light” | Switches between the dark and light theme |
| “help” | Lists what you can say |
If one letter was misheard (“show four paynes”), PaneScale corrects it and runs the command. If it was less certain, it asks Did you mean “…”? instead of guessing.
Privacy
- Audio is recorded only while the panel shows Listening.
- Transcription runs locally with whisper.cpp; no cloud service is involved.
- The temporary audio file is deleted right after transcription. Only the resulting text remains — wherever you dictated it.
See Your data.
Troubleshooting
- Nothing recognized — speak closer to the microphone and try again.
- No recording at all — allow microphone access for desktop apps in Windows Settings → Privacy & security → Microphone.
- Commands don’t work but dictation does — make sure the panel is in Command mode; dictation never triggers commands.