Dictate a task with your voice
Set up local speech recognition, review the transcript, and choose whether to use optional AI polish.
CapabilitiesUse speech as input#
Dictate editable text, review it, then send your task.
Prepare the microphone and model#
Open Settings → Voice in the desktop app. Enable input, grant microphone permission, and select a device. Download a recognition model matching your language and activate it when ready.
Recognition runs locally. Optional AI polish sends transcript text to the configured provider. The first model download needs a network connection.
Dictate a first brief#
- Open the intended conversation and click the microphone. Default shortcuts are Shift+Command+D on Mac and Ctrl+Shift+D on Windows/Linux.
- Speak the short task below.
- Stop recording and wait for transcription and any enabled polish.
- Correct the text before sending.
Read the attached error report.
Summarize reproduction steps and missing information.
Do not modify files yet.
Check and reuse the result#
Success means accurate text appears in the intended composer. Check numbers, paths, and limiting words such as “do not.” A transcript does not mean the task has executed.
History in Voice settings lets you copy transcripts, export recordings, or delete recordings. Audio and text are stored on the device; local recognition does not mean audio is never retained.
Troubleshoot dictation#
- No button: check voice input is enabled in the desktop app.
- No sound: check permission, device, and system input level.
- Failed or poor recognition: confirm the model is ready and suits the language, then try a short recording.
- Polish fails: check its provider and review the available text before sending.
Continue with the instruction#
Attach material through the tools menu. See providers for polish connections and settings and data for data handling.