How Caseo uses AI.
Exactly where it is, and where it isn't. If we can't check it, we don't publish it.
Every catalog photo — heroes, boards, recipe shots — is made to a documented house style, and every new one is checked against a written accuracy rule before it publishes. Each one carries the mark.
The Advisor is a language model, and it can be wrong. Cheeses it recommends are resolved against the Atlas and shown as cards you can open. Every answer is labelled an AI answer.
Name a cheese and the answer comes from 14,000+ curated links, written and checked by the person who built Caseo — no wait, no AI answering you, no charge. Only when the library has nothing does Pair offer the AI Advisor, and it says so before you tap.
Your photos, notes and boards are yours. We don't generate them, alter them, or use them to train anything.
Nothing you type or photograph reaches an AI model until you say yes. You are asked once, the first time a feature would send something, and you can change the answer any time in Settings → Privacy → AI features. Declining does not break the app: the pairing library answers most asks without a model. The check runs on our server at the exact point your words would leave, and if we cannot confirm you agreed, the call does not happen.
We also keep a record of every call we make to a model — which model, when, what triggered it and what came back. That record is anonymous: it is a log of calls, not of people, and it is deliberately not linkable to your account. It never stores your photo, or a fingerprint of it. Neither of these is something we read the law as demanding of an app our size; we do them because we would rather be able to answer the question than assert an answer.
Told us something wrong? Email hello@caseo.app or use the report control in the app — we fix the library, which fixes it for everyone.