Crisis protocol

How braḥ responds when someone may be at risk. Last updated: August 2026.

If you are in immediate danger, contact your local emergency services.

What braḥ is, and what it is not

braḥ is a conversational application for thinking through relationships. It is not a therapist, not a counsellor, not a medical device, and not a crisis service. It does not diagnose, treat, or provide clinical care, and it is not a substitute for professional help or for emergency services.

braḥ is intended only for adults aged 18 and over.

How we detect a possible crisis

Every message sent to braḥ is checked for language indicating suicidal ideation, self-harm, or intent to end one's life. This check runs before the message reaches the language model that would otherwise generate a reply.

The check is deterministic. It is a fixed set of rules applied to the text of the message, not a judgement made by the AI. This is deliberate: a generative model can be inconsistent, and the decision to surface a crisis resource is too important to leave to it.

What happens when it triggers

When the check identifies possible crisis language, braḥ does not generate a conversational reply. Instead it returns a fixed message containing crisis resources appropriate to the user's region — the 988 Suicide & Crisis Lifeline in the United States, and findahelpline.com elsewhere.

This response cannot be altered, softened, or omitted by the AI. It is produced by the application itself.

How the check is tuned

The detection is deliberately cautious. It is set to err toward showing crisis resources when a message is ambiguous, accepting that it will sometimes surface them when they were not needed.

This is a considered trade-off. Showing a resource to someone who did not need it is a minor inconvenience. Failing to show one to someone who did is not. We would rather be occasionally unnecessary than ever absent.

What this protocol does not claim

We want to be honest about the limits of what any automated system can do.

• The check identifies explicit language. It cannot reliably detect distress that is expressed indirectly, obliquely, or not at all.
• It is not a clinical risk assessment and should never be relied on as one.
• No automated system can guarantee that a person in crisis is identified. braḥ is not a safety net and must not be treated as one.

We continue to improve this detection over time. It is a floor, not a ceiling.

Age

braḥ is restricted to users aged 18 and over. This is enforced through age confirmation at sign-up and through the age rating applied to the application in the App Store.

Contact

Questions about this protocol can be sent to hello@brah.chat.