Voice mode
Speech-to-text for Mind CLI chat. Recording happens on your machine; transcription runs on Mind's servers.
Voice mode lets you speak your prompts instead of typing them. Press /voice, speak, press Enter, and the transcript is placed in the prompt for you to review and send.
Your audio is uploaded. Recording is local, but transcription is not. When you stop a recording, the audio clip is sent to Mind's API, which transcribes it with Groq-hosted OpenAI Whisper (whisper-large-v3-turbo). See Privacy for exactly what is sent and kept.
Requirements
- A Mind account. Voice mode calls Mind's transcription endpoint, so it needs the Mind cloud transport. It is unavailable on direct transports — if you have switched to a local model or xAI with
/model,/voicewill tell you so and refuse to start. - A recorder on your PATH. Mind bundles no audio libraries; it shells out to SoX (preferred) or FFmpeg.
- A microphone. There is no remote-microphone option, so voice mode does not work on a cloud desktop without an audio device.
How it works
- Run
/voicein a chat session. Recording starts immediately. - Speak your prompt.
- Press Enter to stop, or Esc to cancel and discard the audio.
- The clip is uploaded and transcribed, and the transcript is inserted into your prompt.
- Review it and press Enter to send.
Nothing is sent to the assistant until you send it. Transcription never submits a turn on your behalf.
Starting voice input
Slash command
Starts one recording in the current session. Running /voice while a recording is already in flight is refused rather than starting a second one — stop the first with Enter.
There is no push-to-talk key binding.
Limits
| Limit | Value | What happens at the limit |
|---|---|---|
| Recording length | 120 seconds | The recorder is stopped automatically and transcribed |
| Clip size | 4 MB | The recording is rejected before upload |
Audio is captured as 16 kHz mono WAV. There is no silence detection — a recording runs until you stop it or it hits the 120-second cap.
Privacy
What leaves your machine. When you press Enter, the WAV clip is base64-encoded and sent to POST /api/agent/transcribe on Mind's API, which forwards it to Groq for transcription with whisper-large-v3-turbo. Cancelling with Esc uploads nothing.
What stays local. The temporary WAV file is deleted from your machine as soon as the request completes, whether it succeeded or failed.
What is recorded. Each transcription is written to your usage ledger with the provider and the audio duration in seconds. Transcription draws no credits; the request counts against your plan's rate window.
What Mind does not do. Audio is not used to train models. Mind does not retain the clip after transcription returns.
If you need transcription that never leaves your machine, voice mode is not currently the right tool — use your operating system's built-in dictation into the prompt instead.
Troubleshooting
| Problem | Fix |
|---|---|
| “No audio recorder found — /voice needs sox or ffmpeg” | Install SoX or FFmpeg (see Requirements) |
| “voice needs the mind cloud transport” | You are on a direct transport. Switch back to a Mind cloud model with /model |
| “Already recording — press Enter to stop it first” | A recording is in flight. Press Enter to finish it, or Esc to discard it |
| “No speech detected — nothing inserted” | The clip transcribed to an empty string. Check your input device and that the recorder captured audio |
| Recording cuts off after two minutes | That is the hard cap. Record in shorter passes |