How Private.Ki uses AI
Private.Ki does not use your emails, chats, attachments or account data with any AI system — ours or anyone else's. Where AI is and is not involved.
Private.Ki does not use your emails, chats, attachments, contacts or account data to train, improve or run AI systems. We do not send your communication data to any AI provider, internal or external. This applies to every Private.Ki service — email and chat, on the web and in the mobile apps.
What AI is never used for
- Your message content. It could not be, even if we wanted to: message bodies, attachments, voice messages and subject lines are encrypted on your device and stored only as ciphertext. There is nothing readable to feed to a model. See What our server can and cannot see.
- Your metadata. The delivery metadata the server does see — who you send to, when, how large — is not shared with AI tools for analysis, summarisation, classification, profiling or automated assistance.
- Spam filtering. Outgoing external mail is checked against DNS blocklists, not scored by a machine-learning classifier. See Third-party services.
- Support. Support requests are read and answered by people.
- Features in the app. The app has no AI writing assistant, smart replies, summaries or automatic categorisation, and no such feature is sending your text anywhere.
Where AI is involved
Writing software. Like most engineering teams, ours uses AI coding tools when writing and reviewing the code that runs Private.Ki. Those tools see source code, never user data or production databases.
Help-center text. Some help-center pages were drafted or summarised with the help of AI, then checked by a person against the app. That work operates only on public documentation — it never involves your mailbox, chats or account.
Nothing else. If we ever add an AI-backed feature to the product, this page will describe exactly what data it sees and how to turn it off before the feature ships.
What our server sees
Cannot see
- Message content or metadata passed to any AI system
- A copy of anything sent to an outside AI provider
Can see
- Nothing additional — this page adds no data collection
Common questions
The help center used to show an AI-generated summary at the top of articles. Where did it go?
The previous help center produced automatic summaries of its own articles. The rebuilt help center does not; each article is written to be short enough to read in full.
Could you train a model on encrypted messages?
No. Ciphertext carries no usable information about its content. And we would not: the point of the design is that we cannot read your messages, so we also cannot process them.
Do the mobile apps use on-device AI, such as keyboard suggestions?
Your phone's keyboard, dictation and autocorrect are features of the operating system, not of Private.Ki, and are governed by Apple's or Google's settings. Private.Ki does not add any on-device model of its own.