What it does,
and what it does not.
WhatsApp bot
In a meeting, on a train, anywhere you cannot press play: forward the voice note and read it instead. No app and no download, and the first 5 minutes need no account. After that it draws on your plan’s minutes like anything else.
There is no list
Nothing here is an integration. The client types into whichever window has focus, so it works in your editor, your mail, your chat, your ticket tracker — and in whatever you install next week, without anyone adding support for it.
Vocabulary, casing and punctuation
Technical, medical and legal vocabulary usually comes through without a word list to maintain and nothing to train. What it cannot learn is a term that exists only inside your practice — an in-house abbreviation, a colleague’s name. We make no accuracy claim; we publish our word error rate instead.
- Casing, punctuation and sentence breaks, applied for you
- Over fifty languages, picked up without a setting to change
Live or uploaded
Dictate live, or upload something you recorded earlier. Same account, same minutes.
- A system-wide hotkey types into whatever window has focus
- Upload finished audio in the browser, no client needed
- One quota covers both — nothing is metered twice
On the paid plans
For recordings with more than one voice in them, and for anyone who has to find a moment again later.
- Speaker diarization: who said which passage (Ultra)
- Word-level timestamps, for subtitling and editing (Pro and Ultra)
Built because typing hurt
This started as one developer’s nerve problem in a left hand, from too many years of typing. Speaking instead of typing removes those keystrokes — and lets you write standing, walking, or looking away from the screen entirely. What that does for a body is a question we treat carefully on its own page rather than in a feature list.
Most people speak faster than they type
Most people type around 40 words a minute and speak around 150. That is a property of human beings, not of this product — which is why the honest thing is to let you check it on your own voice and your own text.
Under the hood
The four decisions behind the behaviour above.
Cut at your pauses, not on a timer
Your text is cut at the pauses you actually make, not on a timer. The silence between them is not sent for transcription, which is why the result reads as sentences rather than as a wall.
Not your laptop’s problem
Transcription happens on our servers, not on your machine. A five-year-old laptop gets the same accuracy and the same speed as a new workstation, and your own hardware barely notices. It does mean you need a connection.
Nothing to integrate
The client writes into the focused field the way a keyboard does, so there is nothing to paste and nothing to integrate. Browser, terminal, editor — the text appears at the cursor.
Deleted as soon as it is transcribed
What happens to your audio is in the privacy policy, not just in the marketing.
- Raw audio is deleted as soon as the transcript exists or the attempt fails
- A transcript from an uploaded file stays in your account for 365 days and can be deleted sooner; dictation through the client is not stored
- Never used to train or fine-tune a model, ours or anyone else’s
- The demo on this site is the exception: its transcript is kept
- Encrypted in transit, processed in Germany
Try it on your own writing
Whether it suits the way you write is something you can only find out by trying it. The free tier is enough for that.
Free tier: 30 minutes a month, no card.
Common questions
Which applications does it type into?+
Usually, because it targets whatever text field has focus rather than integrating with each application. The exceptions are programs that deliberately reject synthetic input — some games, some remote-desktop and virtual-desktop sessions, and some hardened enterprise clients.
Which languages does it handle?+
Over fifty, picked up from what you say rather than set in a menu, so switching language mid-day needs nothing changed. Accuracy is not equal across all of them, and the six we measure and publish are on the word-error-rate page.
Does it work without a connection?+
No. Transcription happens on our servers, which is what keeps a five-year-old laptop as quick as a new one, and it is also why a machine with no network is a machine where this does not work.
What do the paid plans add?+
Longer files, more minutes, and two capabilities the free tier does not carry: word-level timestamps (Pro and Ultra) and speaker diarization (Ultra). The full limits are on the pricing page.
Is there an API?+
Yes — a WebSocket for streaming and a job endpoint for files, with the same account and the same key. The reference lists every parameter and response.
Last updated .