Why on-device AI is the right default
Modern laptops run flagship-class speech models locally. Once that's true, sending your audio to a server becomes a choice — and usually the wrong one.
For years, good transcription meant cloud transcription: the models were too big for consumer machines. That stopped being true. Whisper-class models now run comfortably on an M-series MacBook — Metal does the heavy lifting, and a one-hour interview transcribes in minutes.
The privacy math
Audio is among the most sensitive data you produce: it carries voices, names, and things said off the record. Every upload creates a copy you no longer control, governed by a policy you didn't write and can't audit. Local processing removes the entire category of risk instead of managing it.
It's also simply faster in practice: no upload time, no queue, no rate limits, and it works on a plane. The cloud's remaining advantage — accuracy — evaporated when the big models learned to run locally.
Our commitment
Every app we ship follows the same rule: your data stays on your machine unless you explicitly export it. Not as a premium feature — as the baseline.