Privacy · 27 August 2026 · 4 min read
Why nothing leaves your Mac
On-device speech recognition costs us more to build and ship. Here is the thinking behind doing it anyway.
Most voice tools send your audio to a server. It is cheaper, easier to improve, and it means the company holds a recording of everything you have ever said at your desk.
We did not want to hold that. Murmur AI runs recognition on Apple Silicon, on your machine, and the audio is discarded the moment the words are written.
The practical upsides
It works on a plane. It works in a hospital or a law firm where uploading a patient or client name is simply not allowed. And it is fast, because nothing is waiting on a round trip.
The trade-offs, honestly
The app is bigger to download, because the speech model ships inside it. It needs a recent Mac with an Apple Silicon chip. And we cannot quietly improve accuracy from your recordings, so improvements arrive as updates instead.
We think that is the right side of the trade. Your words are yours.