Hold to Record
Hold a key while you speak, release to transcribe. Fn is bound by default. Native-feeling, no shortcut to remember.
holdfnDictation
Voice to text in any app on your Mac. Pastes at your cursor. No copy and paste, no separate window, no waiting on the cloud.
Other platformsDraft a release email about meetings v2 going GA next Tuesday. Friendly, three short paragraphs, end with a link to the changelog.
text appeared at the cursor in claude.ai
Anywhere a cursor blinks
DuckType pastes at the OS level, not through a browser extension or a per-app integration. If your cursor is in a window and you can type into it, you can talk into it.
AI assistants
Code editors
Writing surfaces
Communication
Browsers and beyond
The list above is illustrative. The actual answer is "any app that accepts keyboard input on macOS." We have not found one where it does not work.
How you start a recording
The right shortcut is the one you forget you are using. Map it to a key your hand already lives near.
Hold a key while you speak, release to transcribe. Fn is bound by default. Native-feeling, no shortcut to remember.
holdfnTap a shortcut once to start, tap again to stop. Good for longer dictation sessions.
tap⌃SpaceDouble-tap a modifier key to start. Quick activation without reaching for a chord.
double⇧⇧Always listening. Voice activity detection chops your speech into utterances and pastes them as you go.
VADLanguage, your way
Most dictation engines output American English by default and call it done. DuckType romanises CJK scripts, corrects to your regional spelling, and learns the words you use that the model has never heard of.
Japanese, Chinese, Korean, both ways.
Get the original script and the Latin transliteration side by side.
British. Swiss. European Portuguese.
Pick the variant. The model output is rewritten to match.
Names. Acronyms. The thing your team renamed last week.
Add words to a dictionary list and DuckType stops getting them wrong. Multiple lists for different contexts.
Skills, chained
What lands at your cursor doesn't have to be the raw transcript. Stack skills on top: clean filler words, translate to French, format as a commit message, write Markdown bullets. They run in one pass.
One pass
um yeah so basically what i'm thinking is we should ship the flag tomorrow and then keep an eye on the dashboards
We should ship the flag tomorrow and then keep an eye on the dashboards.
feat(rollout): ship feature flag tomorrow Monitor dashboards after rollout.
Skills are just prompts you save once and trigger by shortcut. Bring your own model, or use the cloud tier. Local models run for free.
What it does
The opinionated stuff is above. This is what's in the box.
Run a local Whisper, Parakeet, or SenseVoice model on-device. No network, no cloud bill, no audio leaves your Mac.
Plug in OpenAI, Groq, Deepgram, ElevenLabs, Mistral, or Cloudflare keys and pay providers directly. Free with your own key.
Switch language with a shortcut. Mid-sentence code-switching is fine. The transcript holds whatever you actually said.
Misfire? One shortcut deletes the last paste cleanly. No copy buffer pollution, no dangling text.
Every transcript is saved on disk. Search, replay, copy again, or pipe to a script. Local files, plain Markdown.
A ducktype CLI for transcribing files, plus a Claude Code skill so coding agents can listen too. For when typing is the slow path.
Free with local models or your own API key. Cloud transcription is a paid add-on, never required.
Other platforms