Dictate a message
- Open Settings → Preferences → Dictation.
- Choose the Transcription provider and Transcription model for Chat or Code. Each surface keeps its own defaults.
- Confirm the required connection under Models and microphone under Devices, then save settings.
- In the composer, choose Dictate message and speak.
- Finish recording, inspect the transcribed text, and correct names or commands before sending the message.
The normal selected chat or coding model answers the submitted text. Choosing a transcription model does not replace that model or start a voice conversation.
Read a finished answer aloud
- Open Settings → Preferences → Read aloud.
- Choose Speech provider, Speech model, and Voice from the available options.
- Save settings, then use the speaker button on a finished Chat answer or Code turn summary.
- Check the selected Speaker under Devices if playback is silent.
Read aloud is a one-shot text-to-speech request. Changing the live Voice model does not change these preferences. The read-aloud provider choices in this build do not include xAI; choose an offered speech route instead of assuming every voice provider has the same endpoint.
Separate the failure stages
No recorded sound points first to microphone permission or device selection. Recorded sound with failed transcription points to the transcription route or request. Correct text with silent read-aloud playback points to speech generation or output selection.
Use audio devices and voice/audio troubleshooting. Speech requests can have their own recorded usage, described in accounting.