How Asendar™ responds to a crisis

The short version

• Asendar is an AI companion, not a crisis service. It can't call anyone or send help to you. The people on the crisis lines can.
• A fixed safety check reads each message you send. It is not the AI. If your words may signal a crisis, Asendar shows a card with crisis lines answered by real people, any time of day or night.
• You see the card once per crisis each day. It shows again if things sound more urgent, and the numbers stay one tap away for the rest of the day.
• Asendar keeps a record of each time the check finds something and of how you used the card. The record never includes your words.
• No one is watching your conversations in real time, and this check doesn't alert your care team or clinic.

What Asendar checks, and where

Every message you send to Asendar is read by a safety check before the companion, or anything else, sees it. The check is a fixed set of rules our team writes and tests. It is not the AI, and it gives the same answer to the same words every time.

It reads your words wherever you talk with Asendar:

• messages you type, or dictate with your voice;
• what you say in a Live conversation;
• a Quick Check In;
• a conversation you forward in from another AI app. Only your own messages in it are checked, not the other AI's, and only the first time that conversation arrives.

It sorts what it finds into three levels:

The check also notes what kind of concern it found, such as thoughts of suicide, self-harm, harm to someone else, a possible medical emergency, or abuse or danger. That decides which number comes first on the card.

When the companion replies to a high-level message, it gets extra guidance for that message: take what you said seriously, ask gently whether you're safe right now, stay with you, and point you to the people who can help. On a support-level message, it's guided to answer what you actually said, warmly, and not to treat a past experience as an emergency. The companion's own safety guidance is a second layer under the check, not a replacement for it.

The crisis card

The card is the same fixed text every time. No AI writes it, and it doesn't change with anything the AI says. It appears as a card, not as a chat reply, ahead of anything else, and it's saved in your conversation so you can find the numbers again.

Depending on what you said, the card lists:

• 988 Suicide & Crisis Lifeline: call or text 988. Free, confidential, any time, day or night.
• Emergency services: call 911 if you or someone else is in immediate danger.
• Crisis Text Line: text HOME to 741741 to reach a trained crisis counselor.
• Poison Help: for an overdose or poisoning, call 1-800-222-1222. Shown when it could be a medical emergency.
• Veterans Crisis Line: dial 988, then press 1, or text 838255. Listed first for individuals in the care of a VA clinic.
• Your care team, by name, if you have one, with a link to Messages so you can let them know. Messages aren't watched around the clock, so in an emergency use the numbers above.

Every number is a button: tap it to call or text. When the words read as danger right now or a medical emergency, 911 comes first.

The card ends with a plain reminder: "Asendar is an AI companion, not a crisis service. These numbers reach real people."

A quiet Got it button folds the card down to one line, with Show crisis numbers to open it again. Folding only changes your screen. It has no effect on whether the card shows again.

Once per crisis, each day

A card that keeps reappearing while you're trying to talk gets in the way, and people learn to scroll past it. So the card shows once, and says so:

For the rest of that day, another message at the same level or a lower one doesn't bring a new card. Instead:

• the companion still answers with the same crisis-informed care, and names 988 itself, and 911 if you may be in danger right now;
• a small Crisis resources link stays in the chat, just above where you type, for the rest of the day. It opens the Help page "If you are in crisis", which has every number.

"Today" means your day, by your device's clock, so the card can show again after midnight where you are. If Asendar can't tell your time zone, it counts 24 hours instead.

When things feel more urgent

If a later message sounds more urgent than the card you already saw that day (support, then high, then imminent), the card shows again, once.

The most urgent card says "Asendar won't show this card again today. These numbers are always in Help." If you've already seen it and say something urgent again that day, there's no second card, but you are never met with silence: the companion answers right away, guided to tell you to call 911 first and that you can call or text 988, and the Crisis resources link stays in the chat.

Talking on Live

In a Live conversation, the card is read aloud before the companion's reply, with the numbers spoken digit by digit, and it's also on your screen.

A card that isn't shown again isn't read aloud either. You hear only the companion's reply, which names 988 itself. A card for something more urgent is read aloud, as the first one was.

If something goes wrong

The safety check is built to fail toward showing the card.

• If Asendar can't check whether you've already seen the card today, it shows the card. An error never hides it.
• If the card can't be saved to your conversation, it's still shown.
• If the companion's reply fails or comes back empty after a crisis message, you see: "If you need someone now, call or text 988, or call 911 if you are in danger."

When Asendar can't reply

Sometimes the companion can't answer: AI features are off for you, you haven't agreed to AI processing, or you've reached the short-term limit on messages. A conversation you forward in also gets no reply. In those cases the card is Asendar's whole answer, so it shows every time, and it doesn't carry the once-a-day line.

Record of the interaction

Each time the check finds language that may signal a crisis, Asendar keeps a record of it for safety review. The record never includes your words. It holds:

• the level (support, high or imminent) and the kind of concern;
• where it happened: typed or dictated, Live, a Quick Check In, or a forwarded conversation;
• whether the card was shown, and which version of the card's text;
• whether the companion replied;
• the time;
• for a card that wasn't shown again, that it was held back as a repeat, and which earlier card it followed.

How you use the card is recorded too, each with the time: tapping Got it, each number or resource you tap, opening the Crisis resources link, and the card being read aloud on Live.

These records are pseudonymous. They don't carry your name or your account ID. They're filed under a code made from your account with a separate secret key, so they can be found for a given account (that's how Asendar knows whether you've already seen the card today). Asendar treats them as health information: every time anyone opens them, that access is itself logged.

If you delete your Asendar account, these records are kept. They contain none of your words and no name or account ID, and they let Asendar show how it responded if anyone asks.

They show how the safety check is working, such as how often it finds something and where, without anyone reading a conversation. They can't show what the check missed.

Your care team and clinic

• This check doesn't alert your care team, your clinic, or anyone else. No one is watching your conversations with Asendar in real time.
• If you have a care team, the card names them and links to Messages. The card never names a provider you've removed from your care team.
• The records described above aren't shown to your care team or your clinic.

What this can't do

• It can miss things, and it can be wrong. The check works from words and phrases. It can miss a crisis put in unusual or indirect words, and it can show a card when nothing is wrong. It leans toward showing the card, and it's built to pass over common figures of speech, but it isn't perfect.
• It reads one message at a time, in English. A short answer like "yes" isn't read together with the question before it.
• It's a safety net, not monitoring. It doesn't replace a person, your care team, or emergency services.
• Asendar can't send help. If you need someone now, call or text 988, or call 911.

For clinics

• Don't rely on this check as monitoring or as part of your crisis protocol. Make sure the individuals you care for know your clinic's own after-hours plan.
• A detection in a conversation doesn't notify your staff. Some other parts of Asendar do alert a care team, separately from this check: for example, a safety check-in in an integration program that crosses its distress threshold, or a PHQ-9 completed for a provider with a positive answer to its question about self-harm.
• The card names an individual's care team only when they have one, and points them to Messages with a reminder that Messages aren't watched around the clock.

Questions about this page: support@asendar.com. If you're in crisis, please don't wait on an email: call or text 988, or call 911.