Docs
How it reads your data.
Kata sounds like it knows you because it does. Here is exactly how that works, and why none of it means your data lives on our servers.
Assembled on your device
When you send a message, the app builds a compact context block on your device: your phase and goals, recent training, food and protein, sleep, weight history, and Kata's memory of your preferences. That block plus your message is what Kata reasons over.
Sent to answer, not to keep
That context travels encrypted to our AI provider to generate the reply. Kata's servers keep no copy: it is not written to a database on our side, and there is no per-user store of your logs or chats anywhere but your device. The provider processes the request to answer, does not train on it and deletes it within 30 days. The only thing recorded on Kata's servers for a message is an aggregate usage count for fair use, never the content.
Only what the answer needs
The context is a summary, not a data dump. Kata sees enough to be specific about today and this week; it is not handed your entire history for every trivial question. Photos are analyzed the same way, only in the moment you send them, and speech arrives as text: Apple's speech recognition turns it into words on iPhone, your phone's speech recognizer on Android (Google's on most phones, which may process the audio on Google's servers), and your browser's speech recognition on the web.
Because the picture is rebuilt fresh from local data each time, correcting Kata's memory or editing an entry changes what it sees next time, instantly. You are always the editor.