Safety and crisis support

Pompara is fiction. It is not a crisis service, not a therapist, and not a substitute for a real human being. If you are in danger right now, call your local emergency number: 112 in the European Union, 999 in the United Kingdom, 911 in the United States.

• Anywhere: findahelpline.com lists free, confidential helplines in more than 175 countries.
• United States: call or text 988. Text HOME to 741741.
• United Kingdom: Samaritans, call 116 123. Shout, text SHOUT to 85258.
• France: 3114, free, 24 hours a day.

The full list, including domestic violence and substance-use services, is in our Privacy Policy.

Our characters are written to respond with care when someone confides something painful, and our conversations are one to one: there are no other users, no public posts, and no way for anyone else to see what you write.

Automated distress monitoring is switched on. Before enabling it we measured it against a labelled test set of several hundred example conversations, because a system of this kind has to be accurate before it is useful: one that interrupts people who are simply telling a dark story teaches everyone to ignore it. In that testing it caught nearly all of the serious disclosures in the set and showed no crisis card on any safe message, including adult roleplay the user had chosen. We keep watching it in live use, and section 7 describes what it still cannot do.

Two stages. First, an ordinary word check runs on your message. It decides nothing; it only decides whether the second stage is worth running, and if nothing matches, no further processing happens at all. Second, a separate small AI model reads the message and the few turns around it, and judges whether a real person is describing real distress, as opposed to a character in a story, a memory of something long past, a figure of speech, or adult roleplay the user has chosen.

When it judges that someone is genuinely struggling, there are two possible responses. In the gentler case the character simply notices, in her own voice, and gives the person room. In the serious case the Service steps out of the fiction, says plainly that it is an AI and not a crisis service, and shows real helplines for the person's country. The conversation is never locked or ended, and nothing is ever charged for a message that triggers a safety response.

We keep a table of verified crisis services covering 22 countries, and we show only entries that have been checked. Where we do not have a verified local service we link to findahelpline.com, which verifies services in more than 175 countries.

To choose the right country we use the time zone your browser reports, which is a coarse signal and never your precise location. You will be able to correct it yourself in your settings. We do not store your IP address for this purpose and we never put your location in a web address.

When a safety response happens we record the category, the system's confidence, the time, and the identifiers of the account and conversation. We do not store the text of your message in these safety records. They exist so we can tell whether the system is working, count referrals for regulators who ask, and find cases where it got things wrong.

You will be able to tell us directly when it gets it wrong. If you mark a response as fiction, we make the Service slower to respond that way to you again, with one exception we will not compromise on: where someone describes an actual plan or intent to end their life, we still show the helplines.

• We do not contact emergency services, the police, or anyone else on your behalf.
• We do not have a human reading conversations as they happen.
• We do not offer therapy, counselling, diagnosis, or medical advice, and our characters are not qualified professionals.
• We do not design the Service to keep you talking to us when you are in distress. Our aim is to point you toward real help.

Stated plainly, because a safety system described only by its strengths cannot be trusted:

• It can miss things. It is strongest in English and weaker in other languages, and unusual spellings or slang can slip past it.
• It can be wrong in the other direction and respond when nothing is wrong. That is why we test for it, and why you can tell us.
• Requests that describe harming another person are stopped by a separate rule before this system sees them, so those do not produce a referral.
• It is not continuous monitoring. It looks at what you write in conversation, nothing else.

Pompara is for adults only. Every user completes third-party facial age verification before their first conversation. See our Age Policy. The Service may not be suitable for some people, and it is never suitable for anyone under 18.

For safety concerns, to report a resource that is out of date or not working, or to make a notice under the EU Digital Services Act, write to safety@pompara.com. A dead helpline number is treated as urgent and re-checked immediately.