Product docs
Dictation (Beta)
Press a shortcut, speak, and your words land in the focused input field — with voice commands, AI polish, dictation history, and per-plan differences.
Dictation (Beta)
Dictation answers the question "how do I turn speech into text when I don't want to type": press a shortcut, speak, and the result is written straight into whichever input field has focus — no recording pass, no transcription step. It is open to every signed-in user, still in Beta, and both the entry point and the Settings item carry a Beta badge.
Shortcuts and triggers
- Default dictation key:
Fnon macOS,Right Alton Windows - Voice command key:
Fn + Space(Right Alt + Spaceon Windows) - The trigger can be a tap, a hold, or a double-press; releasing a hold does not leak the key to other apps
- A single session can run up to 10 minutes, with a countdown in the final minute
- Enabling dictation on macOS requires Accessibility permission; Settings will say so and can jump you straight to System Settings
Voice commands
Select some text, then say what you need and the AI acts on the selection:
- Rewrite / polish: rewrite the selection the way you asked
- Ask: the answer arrives as a card you can copy, use to replace the selection, or continue inside the AI assistant
- Search: say something like "search … for this" and the browser opens with the query
Voice commands rely on dictation AI, so AI polish needs to stay on.
AI polish
- Two polish strengths: Conservative makes only minimal corrections, Standard rewrites in a written-draft style
- "Polish after dictation" can be switched off entirely, in which case the raw transcript is inserted
- AI polish and voice commands run on a cloud AI service: paid plans use their monthly AI quota, and the Free plan uses the free AI quota provided by the official service
Dictation history
- The Dictation page in the sidebar filters by dictation vs. command and shows counts, words, duration, and speaking rate
- A 7-day and 30-day heatmap, plus a breakdown of which apps your words landed in
- Each record shows the raw transcript, the AI result, and what was actually typed
Plan differences
| Capability | Free | Standard | Pro / Lifetime |
|---|---|---|---|
| Dictation | ✓ (Beta) | ✓ | ✓ |
| Dictation models | Whisper Tiny / Tiny.en / Base / Base.en | All official Whisper models, plus SenseVoice, Moonshine, and the English/Chinese Zipformer models | Every model, including FireRed CTC, Qwen3-ASR, Cohere, Parakeet, and community models |
| Live dictation (text appears as you speak) | ✗ | ✓ | ✓ |
| Saving dictation history | ✗ | ✓ | ✓ |
| AI polish / voice commands | ✓ (uses free AI quota) | ✓ | ✓ |
| Custom instructions | ✗ | ✗ | ✓ |
| Resident engine memory | ✗ | ✗ | ✓ |
- During the trial period your account is treated as Standard, so the Free limits do not apply.
- See the model overview for the full model catalog, and Advanced transcription for per-engine parameters.
FAQ
Q: The shortcut does nothing.
A: Check that the switch under Settings → Dictation is on and a model is selected; on macOS you also need to grant Accessibility permission in System Settings. Then check whether another app has claimed the same shortcut.
Q: I spoke for a while but only got a little text, or it reported a failure.
A: The result is delivered in one go after you stop recording, so long takes wait for as long as they actually ran — the ending is not dropped. If recognition fails the audio is kept, and you can retry it from dictation history instead of saying it again.
Q: Does recognition run locally or in the cloud?
A: Speech recognition runs entirely on your machine. Only when AI polish or a voice command is used does the text get sent to a cloud AI service.
Q: Can I keep dictation results as notes?
A: Standard and above can save dictation history and review it on the Dictation page. When you need to shape the text further, continue in the Note Workspace.
Q: Do exported captions or text carry speaker labels?
A: Dictation itself does not run speaker diarization. After a note transcription finishes, you can run offline speaker diarization and optionally include labels when copying or exporting. See Speaker Diarization.
