Type less.
Say more.
Push-to-talk dictation with AI. It pastes directly where your cursor is. No cloud upload, no account required. About 120 ms from release to text.
How it works
What you say, turned into clean writing by AI
Say 'send an email to my professor' and the dictation becomes a polished email automatically. Pick from Raw · Local · Quick · Think modes.
Give the AI an instruction by voice
Say something like 'summarize this' and the AI runs it on the spot. Beyond dictation — voice as a command.
Questions.
Does it send audio to the cloud?
Never. Whisper runs entirely on your Mac using the Neural Engine (or CPU on older machines). Your audio stays on-device from start to finish. There are no servers, no accounts, no telemetry on the audio itself.
How accurate is it?
Whisper base has strong accuracy for English and most major languages in quiet environments. Accuracy drops with heavy background noise or heavy accents, same as any local model. A future release will let you swap in the small or medium model if you want a precision tradeoff.
Does it work on Intel Macs?
Yes, with a caveat: transcription runs on the CPU instead of the Neural Engine, so latency is higher — roughly 400–800 ms instead of 120 ms. Perfectly usable, just not as snappy.
Can I use it hands-free without holding fn?
Push-to-talk is intentional — it gives you control over when the mic is open and avoids accidental transcription. A toggle mode (click to start, click to stop) is on the roadmap.
Which languages are supported?
Whisper base supports 90+ languages, including English, Korean, Japanese, Spanish, French, German, Mandarin, Arabic, and more. fns does not restrict which languages reach the model — whatever Whisper handles, voice to text handles.
Stop typing the obvious.
Join the waitlist and be first to try on-device voice dictation on your Mac.