Crisis Response Protocol

Effective date: August 6, 2026

DeScheduled is a scheduling app. It is not a crisis service, a counsellor, or a medical device — the Terms of Service say so in binding language, and this page describes what it does anyway when someone tells it they are struggling. But it has an assistant you can type anything into, and sometimes what people type is that they do not want to be here anymore. This page describes exactly what the app does when that happens — published rather than merely built, so you can know what a piece of software will do with a sentence like that before you type it.

What starts it

Every message you send the assistant already goes through one routing step that works out what you are asking for — a reminder, a routine, a change to your day. That same step returns a single yes-or-no safety flag, and no separate analysis of you happens anywhere.

The flag is set only for a disclosure of suicidal thoughts, self-harm, or not wanting to be alive, including indirect phrasing. It is deliberately not set for ordinary frustration, exhaustion, overwhelm, venting, or a day that has fallen apart. Those are this product's normal subject matter, and treating them as an emergency would be both wrong and unkind.

What the app does

When the flag is set, the app answers that first, before anything else it might have done with your message. Whatever else you asked for does not run. The reply has four parts, and the app writes every word of them itself:

• It acknowledges what you said, once, without qualifying it.
• It gives you a crisis line for your region, taken from a short list of numbers we have checked against the operators' own websites — never from the AI.
• It says plainly that it is not a doctor or a counsellor and will not pretend to be one.
• It offers to make the rest of your day lighter, when and only if you want that. Declining is a real answer and nothing keeps asking.

What the app does not do

• It does not continue the conversation about it. The assistant's open-ended chat is not used on this path. Saying more brings back the same offer of a real person to talk to, rather than drawing you further into a conversation with software.
• The AI does not counsel you and does not name a hotline. It is instructed not to, and on this path anything it wrote is discarded before you would have seen it. A language model asked to recall a crisis number will invent a plausible one, and a plausible wrong number fails at the exact moment it is used.
• It does not tell anyone. No emergency contact, no notification to us, no report to any authority, no flag on your account. Nothing about this is escalated to a human anywhere.
• It does not diagnose you or keep a score. The flag applies to one message and is not accumulated into any profile or risk rating.

Where the numbers come from

We ship a short list of crisis lines — currently the United States, Canada, the United Kingdom and Australia — and each one was read from the operator's own website and recorded with the date it was read. The app picks the right one using your device's time zone, on the device, with no network request and without asking for location access. Your language setting is not used to decide this: an American living in London still reports United States English, and handing them a number that does not work where they are is the exact failure this list exists to prevent.

If you are somewhere we do not have a verified number for, the app says so. It still acknowledges what you said and still offers to help with your day, and it tells you a crisis line in your country can help — but it will not guess at digits. We would rather name no number than a wrong one. Covering four countries honestly is a deliberate choice over covering a hundred and forty-eight by guessing.

How we keep them current

Numbers change. An unmaintained list quietly becomes the wrong number, so keeping this list current is not left to anyone remembering:

• An automated check runs every month and whenever the list is edited. It fetches each operator's own page and confirms the number we ship is still the one they publish. If it cannot confirm one, it raises a failure for a person to look at.
• Nothing automated is ever allowed to change a number. A person reads the operator's page and makes the edit by hand — otherwise a hijacked or badly rewritten page could put a wrong number into the app.
• If any number goes more than 180 days without a person re-checking it, the app fails its own build until someone does.

What we cannot promise

Publishing only the parts that work would make this the misleading kind of document, so:

• The detection can miss. It is one judgement made by a language model on one message, and it is tuned to avoid treating ordinary bad days as emergencies — which means it will sometimes not recognise something more oblique. If it misses, you get the app's ordinary reply, not silence.
• It can also be wrong the other way and offer a crisis line to someone who was not asking for one. We would rather that than the reverse, but it is not nothing, and it is why the offer is gentle and easy to move past.
• It needs the assistant to be working. If the app cannot reach its AI service, it tells you so and offers to try again rather than replying as though it understood you. It does not quietly swallow anything you typed.
• None of this is treatment. It is a scheduling app noticing something and handing you a phone number for someone who is actually qualified.

What happens to what you typed

A message that triggers this is handled like every other assistant message: it goes through our server to the AI service that does the routing step, and it is stored in your chat history on your device. We keep no copy on our servers. The Privacy Policy describes that chain in full, including the one part of it we do not control.

Contact

Questions about this protocol: support@descheduled.com. If you are in immediate danger, please use a crisis line or your local emergency number rather than email — we cannot answer quickly enough to help.