Configure and start voice
- Open Settings → Preferences → Voice mode.
- Choose the Chat voice model and Code voice model defaults you need.
- Review the selected provider under How each provider listens. Listening and interruption settings belong to that provider.
- Configure its connection under Models, choose audio devices under Devices, and save settings.
- Open Voice in Chat or Code, review the setup model, and start the session. In Side chat, use Start voice session when available.
- Allow microphone access through the OS prompt if you intend to use the microphone.
The model on the setup pane can override the default before the session starts. A coding CLI subscription is not automatically a voice connection.
During the conversation
Use Mute microphone to pause microphone input and Unmute microphone to resume. End the session with End voice session or the applicable close-voice control when finished. Muting is different from ending a provider connection.
Provider listening settings can change how long a pause is tolerated and how easily speech interrupts an answer. The available controls differ; do not assume OpenAI’s listening choices apply to every provider.
In a Code companion, discuss the current task and verify the context actually available. Voice should not be treated as proof that the companion can see every open file or unrelated application.
If connection or audio fails
Check microphone permission and Devices, then confirm the selected voice model’s connection. A text model that works does not prove a voice endpoint is configured. Use voice/audio troubleshooting.
For speech entered as an editable message or playback of an existing answer, use dictation and read aloud.