How we respond to crisis messages
If you write about suicide or self-harm, the conversation stops being a conversation about thinking patterns. Helplines come first, and nothing stands between you and them.
What happens in the chat
The AI is instructed to recognize messages about suicidal thoughts, self-harm, a wish not to wake up, or violence happening right now. When it recognizes one, it stops the analysis and any exercise in progress, and its reply leads with crisis contacts instead.
These instructions are part of the system prompt in every conversation, in English, Russian and Spanish. The AI is also told what not to say: no "you are not alone," no reassurance that leaves the person where they were.
The safety path is never blocked
- No account needed. A message with explicit crisis wording is never held back by the free message limit. You don't have to sign up to reach this reply, and it doesn't spend one of the free messages.
- It works even when the AI doesn't. If the request to the model fails, your browser recognizes the wording on its own and shows the helplines without waiting.
- Exercises stop too. The same rule applies inside guided exercises and course lessons, not only in open conversation.
Who you can reach
Call or text 988: Suicide & Crisis Lifeline.
Text HOME to 741741: Crisis Text Line.
Call 911 if you're in immediate danger.
All three are available 24/7 in the United States.
These are the lines this site shows. Other language versions of the service show local lines for their own countries.
In the psychological tests
Three questionnaires include items about self-harm and suicidal thoughts: the depression screening, the stressful-event screening, and the wellbeing check. If you answer any of those items above zero, crisis contacts appear on the result screen regardless of your total score. You don't have to land in a "severe" band to see them.
What this service is not
- It isn't a crisis service, and it isn't monitored. Nobody is watching conversations in real time and nobody will be dispatched.
- It isn't a substitute for emergency help, a doctor, or a therapist.
- It isn't a diagnosis. The tests are screening questionnaires, not clinical assessments.
What these checks can't do
The wording check looks for explicit phrasings. Someone in danger may write nothing that matches it, and the AI can miss what a person means. Treat this protocol as a floor, not as protection.
You're talking to software
Helpy is an AI assistant, not a person. It says so when a conversation starts, and the name and status sit above every message. No human reads your messages as you write them.
If you're in danger right now, don't spend time on this page. Call or text 988, or call 911.
This page describes what the service actually does, checked against the code on August 13, 2026.