Keeps everything.
Shares nothing.
An app whose whole point is remembering what you say has to be paranoid about who else could. Here is the architecture, plainly.
Where your data lives
Audio, transcripts, history, and dictionaries are stored locally on your Mac, in an open format you can export at any time. There is no cloud copy of your archive, which also means: we couldn't read it, leak it, or be compelled to hand it over, because we never have it.
Never collected, never trained on
The app never collects your voice or your words, and they are never used to train models — by us or by anyone we work with. The website uses cookie-less, privacy-preserving analytics only.
Three transcription paths
You choose where recognition happens. Managed cloud (trial and Plus): audio is processed in transit and not stored. Your own API keys (Pro): inference runs through providers you choose, on your accounts, skipping our servers entirely. Fully local models (Pro): recognition happens on your Mac; airplane mode works. The default is convenient; the exits are real.
No account system
There is nothing to sign up for. Licenses are cryptographically signed strings verified offline by the app, so activation works without an account, and your identity isn't a database row waiting to be breached. Payments are handled by a PCI-compliant payment processor; card details never touch our systems.
What our servers do hold
The minimum for a paid product to function: license records (a key, a plan, an expiry), device seat counts, and anonymous word-count metering for managed-cloud usage. No voice, no transcripts, no behavioral profiles. Details in the privacy policy.
Reporting a vulnerability
Found something? Email support@acousmos.com with "security" in the subject. Reports go straight to the people who can fix them, and we take them seriously.