Your voice, on your machine.
Dictate prompts straight into the focused terminal or any text field, and control PaneScale with short spoken commands. Transcription runs locally with whisper.cpp. No audio is uploaded, and the recording is deleted the moment it has been transcribed.
- Runs on your machinewhisper.cpp with the model you choose. No cloud, no upload.
- English and GermanFiller words are ignored, near misses get a “did you mean”. Dictation also in Arabic, French and Spanish.
- Nothing risky by voiceDelete, approve, merge, push and deploy always stay a click.
Dictation and commands.
Dictation
Open the voice panel, click the microphone or press Space, speak, and click again. The text lands where your cursor was — a terminal pane, the chat composer, a workflow brief, a task card. If no field can take it, it goes to the clipboard. Up to ten minutes per recording.
Commands
Short phrases that PaneScale understands directly — in English or German. Commands never type into a terminal; they act on the app itself, so “open autopilot” opens the autopilot instead of becoming a prompt.
Commands, at a glance.
Words in brackets are examples. The German equivalents work the same way — for example “zeig sechs Felder” or “was braucht mich”.
| Command | Say | What happens |
|---|---|---|
| Layout | show six panes | Switches the grid to the smallest layout with at least that many panes. |
| Launch | start codex | Opens a new pane with that agent — or “new terminal” for a shell — in the current project. |
| Tonight | tonight update the readme examples | Creates a card marked for tonight’s Overnight Autopilot queue. |
| Task | new task empty state for uploads | Creates a card on the tasks board. |
| Attention | what needs me | Opens the dashboard with every session that is waiting for you. |
| Stop | stop all | Stops running task runs. |
| Open | open autopilot | Opens a mode or place: agent, code, chat, workflow, video, autopilot, settings, tasks, routines, hub, browser, dashboard. |
| Theme | theme | Switches between dark and light. |
| Help | help | Lists what you can say. |
What happens to your voice.
One-time model download
Pick Whisper Base (142 MiB, fast) or Whisper Turbo (547 MiB, more accurate) in Settings. It is downloaded over HTTPS once, checked against a pinned SHA-256 and size, and stored locally.
Local transcription
whisper.cpp runs on your CPU. The binaries come from the official release and are only fetched when you click to install them.
Nothing kept
The recording is written to a temporary file for transcription and deleted right after. Only the text you see remains.
Voice, answered.
Which plan includes Voice?
Local dictation and commands are part of every plan.
How fast is it?
That depends on the model and your CPU. Whisper Base is the quick choice for short prompts; Whisper Turbo is noticeably more accurate and takes a little longer. You can switch any time in Settings → Voice & dictation.
Which languages work?
Both models are multilingual: set a fixed language or let Whisper detect it. Commands are understood in English and German.
Does it need a GPU?
No. whisper.cpp runs on the CPU.
Can I turn it off completely?
Yes. Without a downloaded model, PaneScale never records anything, and the microphone button stays inactive.
Stop typing every prompt.
Download PaneScale, pick a model, and talk to your grid.