Dictation

Whisper Dictation records from the computer microphone and transcribes speech locally. It inserts text into the composer without sending the message.

Set up

1. Open Extensions.
2. Enable Whisper Dictation.
3. Download the multilingual model.
4. Wait for the model to show as verified in the shared Hugging Face cache.
5. Start dictation from the composer and approve microphone access when the operating system asks.

The bundled Extension engine and model weight are separate. Enabling the Extension exposes the action; downloading the model enables local inference. By default the pinned weight lives in the standard Hugging Face hub cache, so Grokship and other compatible software can reuse the same file. Use Settings → Dictation → Model storage to choose a per-app custom cache or restore the Hugging Face default. Switching libraries never moves or deletes a model.

Use dictation

1. Start recording.
2. Speak normally in a supported language.
3. Stop recording.
4. Review and edit the inserted text.
5. Send only when it says what you intend.

Audio remains local to the app. The model remains local to the computer but may be shared by compatible apps. Removing it can make those apps download it again. Silence, very short recordings, or missing microphone permission can prevent transcription.

macOS users can see macOS permissions for recovery steps. Windows and Linux use their native microphone and audio controls.