Use voice
Speak to type in any composer, and listen to replies.
In short
Every text box has a microphone, and every reply can be read aloud. Speech to text can run locally (offline, after a one-time model download) or through Gemini or OpenRouter. Text to speech uses the system voices or Gemini.
1Record
Press the microphone in any composer to start, and again to stop. The words are appended to what you typed, never replacing it. A pill shows the recording time with a Cancel button; recordings are capped at four minutes.
2Pick a provider
Settings → Voice chooses local, Gemini or OpenRouter for speech to text, and the system voices or Gemini for reading aloud. Local speech to text is not available on Intel Macs.
3Listen
Press Listen on a reply, or turn on Auto-read to hear each new reply once.
For technical users
Audio is captured in memory at 16 kHz mono, encoded as WAV and sent once to the chosen provider. Nothing is written to disk unless Keep recordings is on. The window may open only the microphone; a request that includes video is refused. On macOS the permission state is shown with a link to System Settings when denied, and typing keeps working. The local provider is pywhispercpp with ggml-base.en, pinned by SHA-256 and downloaded on first use. OpenRouter speech to text is built and unit-tested but live UNVERIFIED.