Voice

Dictation

Dictation is on-device speech input. Availability depends on the mobile platform and downloaded model.

Live Voice

Live Voice uses one OpenAI API key that you provide. The key is stored on the selected computer, not on the mobile device or hosted control plane. The phone asks the host for a short-lived OpenAI client secret over the encrypted muxr channel.

Audio and realtime events travel directly between the phone and OpenAI over WebRTC. muxr does not proxy the audio.

OpenAI necessarily receives the audio or transcript, the voice prompt, tool names and arguments, and any tool result you choose to return. Use a dedicated OpenAI project key with a restricted budget. Coding agents and commands running as the same OS user may be able to read the host key file.

Settings shows only Configured or Not configured. Replacing and clearing the key happen on that machine; the raw key is never displayed.