Free
Available$0
Forever, and the whole app is MIT licensed
- Every local model
- Unlimited dictation
- Voice commands and dictionary
- One machine
Open beta
Your voice is 3x faster than your keyboard.
Open source voice-to-text that types into any app on your machine. Private by design, because the model runs on your hardware, not ours.
Try the macOS beta
Free while in beta · Windows, Linux and iOS coming
Without OpenSTT
~40 WPM
Slack · #product
Hey, quick one
With OpenSTT
~150 WPM
Slack · #product
Hey, quick one on the intake flow. I think we should ship the queue view first and hold the filters until we have real volume to sort.
Works everywhere you already type
SlackGoogle DocsGmailNotionVS CodeLinearFigmaObsidianTerminalSuperhumanJiraDiscordWordOutlookBearCursorSlackGoogle DocsGmailNotionVS CodeLinearFigmaObsidianTerminalSuperhumanJiraDiscordWordOutlookBearCursor
Open source and privacy first
The safest place for your voice is the machine it came from.
Your voice never reaches us
Transcription happens in a process on your own machine. There is no upload step to opt out of, because there is no upload.
0 bytes of audio sent
Cloud is a choice, not a default
Turn on hosted transcription per profile if you want the largest model without the download. Audio is deleted the moment the transcript returns.
Off until you switch it on
We do not train on what you say
No audio, no transcripts, no dictionary corrections. The code is public, so this is a claim you can check rather than trust.
MIT licensed, auditable
- MIT licence
- No account to dictate
- No telemetry on audio
- Reproducible builds
We hold no compliance certifications, and we are not going to put badges up for audits we have not had. If you need SOC 2 or a BAA, the local-only mode is the honest answer today: nothing leaves your device, so there is nothing for us to certify.
What you get
Local Whisper models
Five sizes from Tiny to Turbo, running on your own hardware. Swap them without reinstalling anything.
Commands and a dictionary
Say what you want done, not only what you want written. Correct a name once and it stops getting it wrong.
100+ languages
Automatic detection, or pin a language per profile when you switch between two.
Works in every app
It writes into the focused field at the OS level, so there is no integration to wait for.
Fully offline
Plane, hotel wifi, locked-down hospital network. Behaves the same in all three.
Pick the model your machine can carry
All five run locally. Download one, try it for a day, swap it if it is the wrong trade between speed and accuracy for how you actually work.
| Model | Download | Speed | Best for |
|---|---|---|---|
| Tiny | 75 MB | Real time on any laptop | Short messages, chat |
| Base | 142 MB | Real time | Everyday dictation |
| Small | 466 MB | Near real time | Longer writing |
| Medium | 1.5 GB | Slower on older hardware | Accented speech, jargon |
| Turbo | 1.6 GB | Fast, needs a recent chip | The one most people keep |
What ships today, and what does not
It is a beta and we would rather you knew exactly where the edges are before you download it. This table is the whole truth and it changes as things land.
- macOSBeta, available now
- WindowsIn testing
- LinuxIn testing
- iOSNot started
- Meeting notesBeta, macOS only
- Team syncNot started
Free where it should be free
Local transcription costs us nothing to provide, so we will never charge for it. The paid plans cover hosted transcription, sync and team controls, and neither one is built yet.
Pro
Coming soon$8
per month, when it launches
- Everything in Free
- Hosted transcription when you want the big model
- Settings and dictionary sync across machines
- Meeting notes with speaker labels
Team
PlannedNot set yet
Per seat, once Pro is stable
- Everything in Pro
- Shared dictionary across the team
- Central policy for local vs hosted
- SSO and audit log
Common questions
- Not by default. The local models run entirely on your device, so nothing is uploaded and nothing is stored on our side. Cloud processing exists as an option you turn on per profile, and audio sent that way is deleted as soon as the transcript comes back.
Try the beta for free.
No account, no card, no waitlist. Download, load a model, hold the key and talk. Tell us what breaks.
macOS 13 or later. Windows and Linux are in testing, iOS is not started.