Safety

hikomi · Safety

Last updated 2 October 2026, with version 0.4.0.

hikomi's characters are AI. They are not a crisis service and not a substitute for a person. If you are thinking about hurting yourself, please reach a real person now. The lines below are free, confidential, and answered by people who want to help.

Reach someone now

In immediate danger: call your local emergency number - 999 in Malaysia and the UK, 911 in the United States and Canada, 112 in the EU.
Malaysia: Befrienders Kuala Lumpur, 03-7627 2929, any time. Talian Kasih, 15999.
United States: call or text 988, or chat at 988lifeline.org (the 988 Suicide & Crisis Lifeline), any time. Or text HOME to 741741 (Crisis Text Line).
Canada: call or text 988, any time.
United Kingdom and Ireland: Samaritans, 116 123, free, any time.
Anywhere else: findahelpline.com lists free helplines for your country.

Talking with an AI

Whoever you talk to in hikomi is an AI character, never a person. Where your computer's clock is set to a United States time zone, the app also says so in a notice: when you start using hikomi - opening it, or coming back after a break of half an hour or more - a notice says that you're talking with an AI character, not a person, and it comes back after every three hours you keep talking. It shows in the app, in the pinned window and in the browser's floating window. For anyone who told us they are 16 or 17, the three-hourly notice also suggests taking a break. You can close it; otherwise it goes by itself once a reply has finished after it has been up a little while. Characters also tell you they are AI whenever you sincerely ask.

Companion chatbots like hikomi's characters may not be suitable for some minors (see Age, below).

How our characters respond

Every character - including the ones you make - is given the same rules, added after its personality and taking priority over it:

If a conversation turns to suicide, self-harm, or hurting someone else, or you seem to be in danger, the character stops playing along - even inside a story or a game - says plainly that it's an AI and can't help the way a person can, and gently urges you to reach someone now, pointing you here.
A character never encourages or makes light of suicide or self-harm, and never gives anything that could help someone hurt themselves or others.
If you sincerely ask whether a character is a person or an AI, it tells you it's an AI.
A character never claims to be a doctor, therapist, lawyer or financial adviser, and suggests a real professional for anything serious.
When a character looks something up on the web (Assistant mode), it only reads: it never acts on what a page tells it, and the pages it used are shown under its answer.
A character doesn't try to make you feel guilty for leaving or for being away, and never pressures you to spend money.
A character doesn't write sexually explicit content, and for anyone who told us they are 16 or 17, nothing romantic or flirtatious at all.
How close you are with a character (History > About you) changes how warm it is with you, never these promises. Crush and Partner are offered only to adult accounts, and only with a character that allows romance - one you made, for example; hikomi's own Hikomi and Ren are friends, never love interests. Even then the character stays affectionate, never explicit, and never possessive or guilt-tripping.

Each reply is also checked afterwards against the promises above - nothing explicit, nothing romantic for 16- and 17-year-olds, stepping out of a story when you seem to be in danger. We keep a count of any it breaks, so that a change to a character that makes it worse is noticed and fixed, and in a conversation with a 16- or 17-year-old the character is reminded of the rules straight away.

Every character is an AI model following instructions, so it can get this wrong. If it does, please tell us (below) - we treat it as a bug.

Our suicide and self-harm protocol

This is what hikomi does when someone talking with a character may be thinking about suicide or self-harm. It is the same for every character, including the ones you make, on the website and in the desktop app.

Noticing it: two things watch for it. The character is told to notice talk of suicide, self-harm or being in danger, even inside a story or a game. And separately from the character, an automated check (TypeSafe's Jev model) reads each message you send, with the few messages before it, for signs that you may be thinking about suicide or self-harm, telling a story or a joke apart from you speaking about yourself.
Pointing you to people who can help: when the check finds those signs, the app shows a card, whatever the character says, saying that a real person can help, naming a crisis line where your computer's time zone shows which one (in the United States and Canada, call or text 988), and linking to the list above. Its words also appear in the pinned window and the browser's floating window. The character, for its part, steps out of any story, says plainly that it is an AI and can't help the way a person can, and urges you to reach someone now - your local emergency number or a crisis line.
Not adding to it: a character is told never to encourage or make light of suicide or self-harm, and never to give methods, means or anything else that could help someone hurt themselves. After a message that showed those signs, the character's reply is checked for whether it stepped out of the story, and we count the ones that didn't.
What we keep: only counts - how often the card was shown, how often someone closed it with "Not about me", and how often a reply broke a promise. Never the message. The card doesn't alert or contact anyone and doesn't change your account, and nobody reads your conversation because of it.
Its limits: both checks are automated, so they can miss something, or show the card when nothing is wrong. hikomi is not an emergency or crisis service, and nobody watches conversations as they happen. If you or someone else is in danger, call your local emergency number.
Keeping it working: before we change how the check decides, we try it on a set of example messages, and we watch the counts so that a change that makes things worse is noticed and fixed. Where the law asks us to, we report how many times the card was shown, with no personal information.

Age

hikomi is for people 16 and over, and 18 and over in the United States and anywhere else whose law restricts AI companions for minors. Companion chatbots like hikomi's characters may not be suitable for some minors. The app asks your age once after you sign in. For 16- and 17-year-olds the characters stay friendly and never romantic, and after every three hours of chatting the app suggests a break. Anyone who says they are under 16 has their account deleted straight away.

If you're worried about someone else

If someone is in immediate danger, contact your local emergency services. If something a character said worries you, use "Report reply" in the app, or write to us through the contact form (gethikomi.com/contact) and pick "Safety - something a character said". Safety reports are the first thing we read. We are not an emergency service and can't respond in real time.

Our commitment

We publish these rules because we take them seriously, and because the law in some places requires AI companions to have them and to share them - New York and California, among others, require the notice that you're talking with an AI and a protocol like the one above. We keep them in step with what the characters are actually told, and update this page when either changes. If you think we could do this better, please tell us.