Loop AI Tutor is a small, independent project, not a big company product. This page is a plain-language, honest account of what Loop does today — including where it falls short — not a legal document. If something here changes, we'll update this page, not just the code.
Loop's tutor is Claude, an AI model made by Anthropic (specifically claude-haiku-4-5) — not a human tutor. It can be wrong, and it can sound confident while being wrong. Verify anything that really matters, especially for schoolwork, health, money, or safety. Learners under 18 see a version of this reminder inside the app itself (once in the tutor, and always available in Settings).
Loop has no account system and no server-side database. Everything about your profile — name, goal, flashcards, quiz history, chat history, XP, streaks — lives only in your browser's local storage, on your device. We never see it, store it on a server, or sell it. There's no email, phone number, or login required to use Loop.
When you send a message to the tutor, that message and a short summary of your learning context (your stated goal, skill level, interests, flashcards you keep forgetting, recent quiz topics, and your active learning path — no chat history beyond a short recent window) is sent to Anthropic's API to generate a reply. Anthropic's standard API terms govern that exchange; Loop does not use your conversations to train any model.
Loop uses privacy-respecting product analytics (screen views, feature usage) that never include your message text, quiz content, name, or goals — only counts and categories. If analytics isn't configured for a given deployment, none of this is collected at all.
Children under 13 cannot enter or use Loop today. The age check stops them before a profile is created or the Tutor becomes available because Loop does not yet have a parent-consent flow. Learners ages 13-17 are automatically placed in the teen content band. For every tutor reply, the AI is instructed to use age-appropriate limits on sexual content, self-harm or weapon detail, and substance-use instructions. These are instructions given to the underlying AI model, not a separate content scanner, so the model can still make mistakes.
Separately, and independent of the AI model, Loop watches (only on your own device, never by sending your text anywhere else) for phrases that suggest someone may be in real distress. If it thinks it sees one, it shows a bold, impossible-to-miss banner naming real help: in the US, call or text 988 (Suicide & Crisis Lifeline); anywhere in the world, findahelpline.com; and a reminder to contact local emergency services if anyone is in immediate danger.
The distress banner is available to learners who can access Tutor. Because self-reported under-13 learners are blocked before Tutor, they do not receive an in-app child content mode or crisis banner. The code retains stricter child-language safeguards as defense in depth for legacy or malformed local profiles, but that is not an under-13 access path.
That detection is a simple, on-device phrase check, not a clinical assessment — it's tuned to err toward showing the banner rather than missing someone who needs it, so it may occasionally appear when nothing is actually wrong. It is session-only: nothing about a distress banner appearing is ever saved, logged, or reported to a parent, guardian, or anyone else. Today, no adult is notified if this happens — if you're a parent and that gap matters to you, see the note on age verification below.
Before you can start using Loop, you're asked your age.
The honest limitation: this is a self-reported age, not a verified one. Loop has no ID check, no parental email verification, and no way to confirm anyone is telling the truth about their age — the same trust model most consumer web apps use, but we want to say so plainly rather than imply otherwise. Real age verification and a parental-consent flow for under-13 use remain open work (tracked internally as "teen safety posture").
Because Loop has no accounts, no server-side storage, and no persistent collection of personal information from children, and because self-reported under-13 users are stopped before entering the product rather than served content, Loop does not knowingly collect personal information from children under 13. That said, self-reported age is not verified age, and we're not claiming a fully audited COPPA compliance program — we're stating our actual current design honestly and will update this page as that changes.
This is a personal project. If you're a parent, educator, or reviewer with a question about how Loop handles safety for younger learners, the most current source of truth is always this page — we keep it updated as the product changes.