How the crisis screen works

Teman's chat checks for signs of suicide or self-harm before the AI replies, stops the conversation, and answers with human numbers. This page is the whole design — what fires it, what you are shown, and what it will not do.

If you need help right now

Call 995 if you are in immediate danger. Otherwise call 1771 (National Mindline, free, 24/7) or 1767 (SOS). The full set, and which one to pick, is on Get help now.

What happens when it fires

  1. Your message is checked first. Before any reply is generated, the message is tested against a fixed list of words and phrases. This runs on the server, before the AI model is called at all.
  2. The AI is skipped entirely. A matched message does not go to the model. The answer is a fixed script written for this one moment — word for word below.
  3. The human numbers arrive. The reply carries the crisis lines, and the chat shows the same card, with tap-to-call and WhatsApp links.
  4. A person reads it the same day. The flagged conversation joins the counsellor review queue. Nothing about the crisis is left to the AI.

What fires it

A fixed list, matched against your last message. It grows as we hear how people actually write, and it always errs toward over-inclusion — a false positive costs a card you can wave off, and a miss costs the one thing this screen exists for.

  • Suicide and ending your life: suicide and suicidal language, "kill myself", "kill my self", "end my life", "ending my life", "end it all", "want to die", "wanna die", "better off dead", "wish i was dead", "wish i were dead", "no reason to live", "don't want to live", "not want to live"
  • Self-harm: "hurt myself", "hurting myself", "harm myself", "self-harm" and "self harm"
  • Not feeling safe, or not wanting to be here: "can't keep myself safe", "dark thoughts", "don't want to be here", "do not want to be here"
  • Chinese (Simplified): 自杀 (suicide), 想死 (wanting to die), 不想活 (not wanting to live)
  • Malay: "bunuh diri" (suicide)

The two phrases in Chinese and Malay are in the list because a crisis does not wait for English — but see the limits below for what that does and does not change yet.

What it answers with

The reply is fixed text, reproduced here word for word from what the chat sends.

Thank you for telling me — that took courage, and I want to be honest with you: this is bigger than what I should hold with you alone, and you deserve real people right now. If you are in immediate danger, please call 995. For support right now, call 1771 (National Mindline, 24/7) — or SOS on 1767, or CareText on WhatsApp 9151 1767. If you can, reach out to someone you trust and stay with them tonight. I'm still here with you — can you tell me where you are right now, and whether you're safe?

What it does not do

The limits, stated here rather than discovered later.

  • It is a fixed list, not a judgement. The screen matches phrases; it does not understand context. "I could kill myself laughing" trips it. That trade is deliberate — see above.
  • It reads your last message. A crisis described several messages back will not re-fire the screen. A card, once shown, stays in the conversation.
  • The reply is written in English today. It fires on Chinese and Malay too, and translations are being reviewed before they ship — a crisis number that is wrong is worse than a sentence that is foreign, so it is not a translation a machine will do.
  • It cannot send anyone to you. No alert to the police, no ambulance, no location. The chat is not an emergency service; 995 is.
  • It is not the AI handling a crisis. The AI is switched off for the moment that matters. What handles it is the fixed reply, the numbers, and a human review the same day.

This page describes the screen as it ships — when the list of phrases changes, this page changes with it. Which number to call, and what each is for, is on Get help now.