OpenSSDocumentation
DesktopLocalApple Silicon

Desktop voice input

Record and transcribe speech locally, then review or submit the resulting text.

Updated 22 July 2026Applies to 0.2.8
On this pageVoice flowUse voice inputRecognition limitsVoice privacy

Voice flow

Terminal
Microphone → Apple on-device transcription → composer → existing chat route

OpenSS uses AVFoundation for audio capture and Apple Speech APIs for on-device recognition. Raw audio is not sent to OpenSS, OpenAI or Anthropic.

Use voice input

  1. 1
    Start

    Select the microphone button. OpenSS requests Microphone and Speech Recognition permissions if needed.

  2. 2
    Speak

    A visible listening state, elapsed time and partial transcript remain on screen.

  3. 3
    Finish

    Stop manually or pause long enough for end-of-speech detection.

  4. 4
    Review

    The final transcript is placed in the composer. Automatic sending waits briefly and is controlled by the voice auto-send setting; editing cancels auto-submit.

Recognition limits

  • On-device recognition is required; OpenSS does not silently fall back to cloud transcription.
  • Silence detection is roughly two seconds, with a 90-second maximum recording.
  • Low-confidence text remains in the composer for manual correction.
  • Temporary audio is cleaned after success, cancellation, error and interrupted-session cleanup.

Voice privacy