Every voice app makes one decision before anything else: where does the audio go? Most send it to a server, because a server is easy to bill for. Psst runs the model on your Mac. Here is what that changes.
1. The audio never leaves the room
A recording of you thinking out loud is one of the most personal files you can make. It has names, numbers, half-formed ideas, the tone of your voice. When the model runs on your Mac, that file is written to your disk and nowhere else. There is no upload, no retention policy to read, no "we may use your data to improve our service". The question "who can hear this?" has a one-word answer.
2. It costs nothing per hour
Cloud transcription is priced by the minute. A rough shape of the market today: about $0.36 an hour for OpenAI's Whisper API, about $0.40 an hour for ElevenLabs Scribe, and subscriptions of $8 to $15 a month for the dictation apps built on top of them. If you talk for two hours a day, that is a real bill. A model on your Mac costs electricity. The download happens once.
3. It is faster than the network
On Apple silicon, Whisper runs on the Neural Engine and the GPU. A large-v3-turbo model transcribes a minute of speech in a few seconds on an M-series Mac, and there is no round trip to a data centre. Words appear while you are still talking.
4. It works on a plane
Local means offline. Wifi off, aeroplane mode, a hotel with a captive portal: none of it matters. Psst keeps writing.
5. The files are yours
Psst saves every transcript as a plain text file in a folder you choose. Spotlight indexes it. Any editor opens it. Time Machine backs it up. There is no export button because there is nothing to export from.
6. No account
An account exists so a company can meter you. A local app does not need to know who you are. You download it, open it, and it works. The licence is a key you paste once, checked once, then never again.
When the cloud is still the right call
Be fair to the other side. A cloud engine can be better at a hard recording: a language your local model does not know well, four people talking over each other, a bad microphone in a loud café. That is why Psst lets you paste your own ElevenLabs or OpenAI key and switch engines per recording. Your key, your bill, your choice, and the default stays local.