Skip to main content
Dictation turns speech into text in the app you are using. Voice Trigger / Voice Command sends a spoken request to an Agent. Live Voice is a separate two-way conversation feature.

​
Dictate into an app

  1. Open Settings → Dictation → Dictation Models and configure a provider.
  2. Open Dictation settings and check the hotkey and gesture mode.
  3. Place the cursor in a text field in another app.
  4. Use the configured gesture to start recording, speak, and stop according to that gesture’s mode.
  5. Review the inserted text, especially names, dates, and numbers.
Microphone permission is needed to capture audio. Inserting text into another app can require Accessibility permission.

​
Post-actions

Dictation can run configured commands after recognition, such as Dictation Polish or Dictation Translate from Writing Toolkit. Review the post-action selection when the inserted text differs from the raw transcript. A post-action using a cloud model can send recognized text to that provider even if recognition itself was local.

​
Send speech to an Agent

  1. Open Settings → Agents → Voice Trigger.
  2. Configure Triggered Agent, its hotkey, and Session Timeout.
  3. Invoke the gesture and speak a complete instruction.
  4. End recording and review the Agent’s progress and result.
Voice Command can reuse the latest triggered Agent session within the configured timeout. It does not work by matching only a fixed set of phrases; the selected Agent interprets the request with its available tools.

​
History and vocabulary

Open Dictation History to revisit past recognized text. Use Dictionary for custom vocabulary and provider language/hot-word settings when supported.

​
Choose recognition separately

Dictation, recording, file transcription, and live captions have separate model settings. A provider suited to batch files may not be the best choice for streaming input. The Speech Recognition guide and provider catalog explain those categories.

​
Troubleshooting

Use Voice Notes to save speech directly into a knowledge base.