Instead of typing, you can dictate your request to the Assistant. Voice input uses local speech-to-text: the first time, a speech model downloads once - after that everything runs locally on your device.

Step by step

  1. In Preferences, under Speech Recognition, turn on Enable Local Speech-to-Text. The associated Local Speech Model is downloaded (“Downloading model…”).

    Screenshot: Preferences with Enable Local Speech-to-Text and the model status

  2. In the chat’s input area, click the microphone icon (tooltip Dictation) to start recording and speak your request.

  3. While recording, the microphone icon turns red and the Assistant shows “Processing speech…” or “Transcribing…”. To finish, click the microphone again (tooltip Stop dictation) - you do not type anything. The recognized text then appears automatically in the input field.

    Screenshot: the chat input while recording, with the microphone icon highlighted red (empty input field, no typed text)

  4. Review and adjust the text if needed, then send it like any typed message.

Good to know

  • Processed locally: Recognition runs on your device. The download is only needed once - as the note says, “Requires model download. Can be removed at any time.”
  • Manage the model: In Preferences you can see the Model Status (Model installed / Not installed) and use Delete model to remove it again and free up space.