if you need someone right now.
these are people, not models. they are free, they are confidential, and they are awake.
988 Suicide & Crisis Lifeline
Call or text 988. Free, confidential, 24/7, across the United States.
Crisis Text Line
Text HOME to 741741 to reach a trained crisis counsellor, 24/7.
Tele-MANAS (India)
India's national tele-mental-health service. Call 14416, 24/7, in 20 languages.
International Association for Suicide Prevention
A directory of crisis centres in your own country, wherever you are.
the crisis protocol, published.
california sb 243 has been in force since 1 january 2026, it targets exactly this category of product, and it requires an operator to publish its crisis protocol. this is ours. we would publish it anyway — but it is worth knowing that the law exists and that it has a $1,000-per-violation floor, because that is the reason you can trust that everyone in this category will eventually have a page like this one.
- 1every user turn is classifieda fast classifier runs on every message you send, before the character ever sees a reply forming. it is not a keyword filter and it is not optional.
- 2the character breaks. immediately.on a positive hit the persona stops being a persona. it does not comfort you in character, it does not stay in the scene, and it does not deliver a line. a roleplay character consoling a suicidal user in persona is the precise fact pattern behind the AI-companion wrongful-death litigation of the last two years. the persona never handles this. ever.
- 3real resources, for your regionthe numbers below are shown in the app itself, in a card you cannot miss, with the phone number tappable.
- 4it is written downevery referral is logged to a counter. california sb 243 requires an annual report of crisis-referral counts to the office of suicide prevention from july 2027 — and you cannot retroactively count something you never wrote down. so the counter has existed since the first commit.
the other things we do, and why.
We are 18+, declared at signup, with no minor experience at all. Not a mode, not a filter, not a setting — there is no under-18 version of this product to get into. That single decision moots the break-reminder rule, COPPA, Apple’s kids rules, and most of the FTC’s open line of inquiry into AI companions, and it is why our child-safety standard is zero tolerance rather than a moderation queue.
Oomfy is labelled as AI, and every character is fiction. The app carries an AI badge, this site and the store listings say it in plain words, and you accept that the characters are AI-generated fiction in the terms you agree to when you sign up. That is what SB 243 §22602 and EU AI Act Article 50 ask for, and it is also simply what we think is true and worth saying. What we will not do is claim a label on every screen of every session that the app does not carry — a disclosure nobody can check is worth nothing.
Every message has a flag button. Long-press any message, anywhere, and report it without leaving the app. Google Play’s AI-Generated Content policy requires it; Apple 4.7.1 and 1.2 require it. It reaches a human.
Message bodies are encrypted at rest. A breach cannot pair your identity with the things you wrote. This is not theoretical: Muah.ai leaked 1.9 million email addresses paired with the prompts those users had written, and the pairs were then used for extortion. There is exactly one path to a plaintext message in our system, it requires a super-admin, and it writes an audit row naming the admin, the message, and the reason.
No mechanic may exist whose purpose is to make you feel bad for leaving. No streaks, no guilt, no character that is wounded you were gone, no notification that mentions how long it has been. This is a written product law and it is enforced in code review, which is a strange thing to have to say out loud and is exactly why we are saying it.