Say it, and it's already written
Click the mic and say what's on your mind. Words appear as you speak; when you stop, the fillers, repeats and misheard names are already sorted and the punctuation is in, so it's ready to send from the composer. Correct a word once and Klaus remembers it.
so um move Friday's review Move Friday's review to to Monday, and loop in EvaAoifeDictionary too.
can you Can you uh change the the login button color to brand red, then run the store frontstorefrontDictionary tests?
okay the PR feedback is The PR feedback is is all addressed, so like merge it into mainemain when the checks pass.
Tidied, never rewritten
Tidying does only what makes the text ready to send. It never expands, answers, translates or summarises, and your meaning and tone stay exactly as you said them.
Punctuation added
lets stop here today and pick it up tomorrow
Let's stop here today and pick it up tomorrow.
Fillers removed
um so I think that works
I think that works
Repeats removed
move it to to Monday
move it to Monday
Names from your dictionary
send it to Eva to check
send it to Aoife to check
Correct it once, and it remembers
After the text lands in the composer, Klaus watches the edits you make next. Genuine corrections are learned: misheard names and terms go into the dictionary, and the way you like to phrase things goes into personalization.
- 1
You fix it onceChange the misheard “Eva” to “Aoife” and send as usual
- 2
It joins the dictionary“Added to dictionary: Aoife” appears, and one click undoes it
- 3
Right the next timeSay the name again and tidying spells it the dictionary's way
- Dictionary
- Add and search words, filter by auto-added or manually added; import up to 3,000 words from one CSV
- Personalization
- Learns your phrasing from your corrections; turn it off any time and the dictionary keeps working
- Only real corrections
- Untouched, cleared or wholly rewritten drafts are never learned as corrections
- Transcription history
- Kept on this computer for 30 days by default, so you can tell a mishearing from a tidying change
Your own recognition source
Choose it in Settings → Providers → Capability settings → Speech recognition: system speech recognition on macOS, a subscription account you're signed in to, or a transcription endpoint you set up. If the source is unavailable you're asked to reconnect; it never quietly switches to another account.
- System speech recognitionmacOS
QoderSubscription
QoderWorkSubscription
QwenWorkSubscription- Custom transcription endpointOpenAI-compatible
Questions
Is it sent as soon as I stop talking?
No. The recognised, tidied text only goes into the composer, where you can edit it like anything you typed; whether and when to send it is up to you.
Which model does the tidying, and does it use my quota?
Live sources that tidy natively return the tidied result in the same speech session. Otherwise it uses the global low tier in Default models, or the chat's main model when none is set, and counts against that service's usage. If the model is unavailable or the request fails, you keep the original transcript.
Where do my voice and text go?
Audio goes to the recognition source you chose. For tidying, the transcript goes with your dictionary, phrasing preferences and a little recent conversation to the source or model doing the tidying. The dictionary and transcription history stay on this computer.
Does it work on Windows?
Yes. Pick a signed-in subscription account that supports speech recognition, or a transcription endpoint you set up, as the source; system speech recognition is currently macOS only.
How long can one recording be?
Up to 5 minutes. You can also start and stop with ⌘/Ctrl+⇧V instead of clicking the mic.