Voice dictation

Prerequisites

• The widget is open. See For clients.
• The browser can record audio/webm with Opus, or audio/mp4, and it exposes the microphone.

Steps

1. Press the microphone. Its name is Dictate. In French it is Dicter. In Spanish it is Dictar.
2. Speak. Recording stops at 2 minutes.
3. While the audio is sent, the box says Transcribing… (Transcription… / Transcripción…).
4. The text is added to the draft. Read it, edit it, then press Send. Dictation does not send the request.

The recording is not stored. Only the text is added to the draft.

Transcription is included. Its cost is counted in AI spend this month. Nothing else to configure.

Verify

The spoken words appear in the text box, and the request is still unsent until you press Send.

On the project overview, AI spend this month includes this cost. Dictation does not deduct a credit by itself. Credits are charged when an agent run finishes. See Autopilot and limits.

Troubleshooting

• The microphone button is missing. The browser cannot record, or it has no microphone API. Typing still works. The button stays hidden in that case.
• "Allow the microphone in your browser to dictate." The site was blocked from the microphone. Allow it for this site, then press Dictate again. See MDF-WIDGET-004.
• "Transcription failed. Try again." Type the request, or try again. See MDF-WIDGET-002.
• Each client can transcribe 30 times per hour on a project. Past that, the widget shows the same "Transcription failed. Try again." message.