Local by nature.
Private by consequence.
There is no server between you and your words. Recognition happens inside your Mac, the microphone closes, and the text is already where your cursor was. Nothing to upload, nothing to delete from someone else’s disk.
Adopted by people who read the source before installing:
Nothing to opt out of
Consent screens exist because something is being collected. Here the pipeline ends on your machine: the dictionary sits in your user folder, the history is a file you can open, and both disappear the moment you say so.
Check it, don’t trust it
Open the code, open the network monitor: both stay quiet.
See how it is builtReadable line by line
The engine is published: clone it, compile it, search it for a single outbound request. Several people already did, and wrote about it.
Compliant by shape
With no processing on our side there are no transfers to map, no retention clocks and no sub-processors to chase. Business gets the paperwork anyway.
Locked to the device
Audio and transcripts stay inside the machine that produced them. On Business an admin can bolt that behaviour down for everyone.
Kept on your disk, or not kept at all.
Every optional setting starts off, and dictation behaves exactly the same with all of them off.
Works with the cable out
No sign-up to start writing. Turn off Wi-Fi and every feature keeps behaving, because none of them needed the network.
You set the memory
Choose how long history survives — a week, a session, or nothing at all — and empty it whenever you feel like it.
Your voice is not fuel
Recordings are never used to improve anything of ours. There is no telemetry channel to disable, because none was built.
What people ask before installing
Where does recognition actually happen?
Inside the app, on your own CPU and Neural Engine. With no inference server in the loop there is nothing to intercept and nobody to subpoena.
What ends up written on my Mac?
Two plain files in your user library: the dictionary and, if you keep it, the history. Open them, move them, throw them away.
Is anything ever sold or handed over?
There is nothing to hand over. Your audio and text never reach us, so no request, buyer or partner could ever be served. Subscriptions pay for the work.
Does my voice end up in a training set?
No. The model improves between releases through public datasets and our own recordings, then ships to you as a version bump.
How do I confirm this on my own?
Build the engine from source, then dictate for ten minutes with Little Snitch or a packet sniffer open. The silence is the proof.
What does an admin get for a team?
On Business: single sign-on, seats and roles, a shared dictionary, and one switch that pins on-device processing for every member.
Do you sign agreements for companies?
Yes, with Business — and it reads quickly, since the only data we process is what your finance team sends us to invoice.
What stays behind if I uninstall?
Only free disk space. Drag the app to the bin and the dictionary and history go with it; no shadow copy exists anywhere else.
Speak freely. Nothing leaves.
Hold the shortcut, talk, keep writing. The words land in the app you were already using and stop there.
Free tier with no expiry · no sign-up · nothing uploaded, ever