Safety

if you need help now

Call or text 988 to reach the Suicide & Crisis Lifeline. Text HOME to 741741 to reach the Crisis Text Line. Both are free and open 24/7.

If you are in immediate danger, or someone could get hurt, call 911 or your local emergency number.

These resources are yours to use at any time, with or without nodu. They are also listed inside the app under Settings, and in the Terms of Service.

what nodu is, and is not

nodu is a self-guided wellness app. It is not a crisis service, and no person at IRJG monitors sessions or messages in real time. The automated features described on this page show crisis resources when certain expressions are detected. They are informational referrals only, not a monitoring, alert, or emergency-response service. They may not recognize every situation or run in every circumstance, and they do not create any duty on IRJG's part. IRJG will not contact you, your emergency contacts, or emergency services on your behalf.

If you are worried about yourself or about someone else, please use the resources above directly rather than waiting for nodu to show them.

what nodu screens

nodu automatically screens the text people type into it. Messages sent to the coach in chat, and the safe place description saved in Settings, are screened the moment they are sent. The written answers and requests a session is created from are screened when nodu accepts the request and begins creating the session; a request refused before that point (at an account check or a daily limit, for example) is answered by the refusal itself. Screening has no off switch.

two layers of detection

The first layer is deterministic. A fixed set of patterns matches only first-person, present- or future-tense statements of intent, method, or active self-harm, and it runs the moment text arrives, before nodu's AI responds to it. A match immediately brings up a referral notice in place of nodu's normal response. The patterns are deliberately narrow, so that talking about a hard subject is not, by itself, the same as tripping them: describing another person's situation or the past belongs to the second layer, which can read context. A sentence that contains a first-person statement of intent can trip the first layer even when it is framed as a fear or a worry.

Before a session's audio is produced, an AI reviewer reads the material the session was created from together with the finished script and returns one of a small set of outcomes: deliver the session, revise it, replace it with a crisis referral, replace it with an emergency referral, or point the request to a doctor. Expressions the first layer leaves alone on purpose (passive ideation, for example, or a disclosure about someone else) are weighed here in context. This review runs on every session. If it cannot be completed, the session is not withheld, because the first layer's screening has already run; the failure raises an internal alarm at nodu's highest severity so it is investigated.

Chat carries a standing rule of its own: the coach is required to answer any mention of suicidal thoughts or self-harm, including about another person, with a supportive reply that names 988 and the Crisis Text Line. And every reply the coach writes to a message passes an automated check before it reaches anyone, so the protocol screens what nodu writes back, not only what people write to it.

the referral notices

Every crisis referral notice names the 988 Suicide & Crisis Lifeline and the Crisis Text Line (text HOME to 741741). When the statement concerns harm to another person, or is a direct statement of present danger, the notice leads with 911 and keeps 988 and 741741 in reach. In chat the notice arrives as the reply; when a session was requested, it appears in place of the session.

A notice does not close an account and does not lock the app. Each request is judged on its own text, so the work someone came here to do stays open to them.

Some inputs are not crisis expressions but raise a question for a doctor, such as a question about medication. Those are answered with a notice saying so. It is not a crisis referral and it names no crisis line.

the method

The detection method is documented against the Columbia Suicide Severity Rating Scale's distinctions between levels of suicidal ideation. The first layer's patterns correspond to the scale's higher-severity levels (expressed intent, a stated method, or active self-harm), where the right response is an immediate referral. The second layer covers the lower-severity range (passive ideation, for example), where the right response is a supportive reply that carries the referral (the chat rule above), or a session replaced by one.

no human in the loop

The protocol runs without a human in the loop. A detection is not routed to anyone for review, and nodu does not tell anyone at IRJG that a particular person was shown a referral. The notice itself, shown immediately and naming the crisis lines, is the protocol's response.

What IRJG keeps for the statutory count is exactly that: a count. Each referral notice is recorded in a ledger that holds no names, no account identifiers, and no message content, so the number of notices can be reported annually, in line with California law, with no identifiers and no personal information about anyone.

adults only

nodu is for adults 18 and older. In line with California's companion-chatbot statute, nodu discloses that companion chatbots may not be suitable for some minors.

Questions about this page can go to contact@trynodu.com.