Windows Codex app dictation records audio but silently produces no transcript

Open 💬 2 comments Opened May 5, 2026 by two3pro

What version of the Codex App are you using?

Codex App 26.429.3425.0

What platform is your computer?

Windows x64

What issue are you seeing?

In the Windows Codex desktop app, dictation can start and appears to record microphone input, but stopping or submitting the recording produces no transcript and no visible error.

The microphone permission is allowed by Windows privacy settings and the security log shows Codex.exe microphone access was allowed. The dictation UI shows waveform activity and a recording timer, so audio input appears to be captured. However, after clicking the stop button or the submit/send button, no text is inserted into the composer and nothing is sent.

The same workflow works on the Mac Codex app, where voice input is transcribed and submitted automatically.

What steps can reproduce the bug?

  1. Open Codex App on Windows.
  2. Open any chat.
  3. Click the microphone/dictation button in the composer.
  4. Speak for several seconds.
  5. Observe that waveform activity and the timer are visible.
  6. Click stop, or click the send/submit button.
  7. Observe that no transcript is inserted and no error is shown.

Expected behavior

After stopping/submitting dictation, the recorded audio should be transcribed and inserted into the composer or sent, matching the Mac app behavior. If transcription fails, Codex should show an error instead of silently dropping the recording.

Additional observations

Local inspection suggests this may be a Windows/Electron MediaRecorder timing issue. The composer dictation path uses MediaRecorder and appears to return early if the recorded chunks array is empty. On Windows, the final dataavailable event may not have populated chunks before the stop handler checks it, causing the recording to be treated as empty and silently discarded.

Related but different issue: #13541 reports dictation failing to start. In this case dictation starts, waveform/timer are visible, but transcription never appears.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗