Mental health · built in, not bolted on
Help that arrives before you ask twice.
A student ten thousand kilometres from home should never have to work out who to tell. One button, a named human, and a clock that does not stop at five.
A student ten thousand kilometres from home should never have to work out who to tell. One button, a named human, and a clock that does not stop at five.
“The measure of a programme is not how few students struggle. It is how fast the right person finds out.”
The principle every part of this page is built on.
No triage menu. No “choose a category”. A student in trouble does not have the composure to classify their own crisis, so we never ask them to.
The button sits in the same place across the student portal. Sending a request requires a connection; when the network is unavailable, the interface must say so and the local emergency number remains the first route for immediate danger.
Pressing it creates a priority support request. The current owner and the organisation’s support rota are alerted with the student, placement and request context. It is not a replacement for emergency services.
Most of it is not a crisis. It is week six, and nobody noticed.
Homesickness rarely announces itself. It shows up as three quiet check-ins in a row — which is exactly the shape we watch for.
A wellbeing check-in is not a game. There is no streak to break, no badge to earn, and a low score is never a failure — it is simply information reaching the right person early.
One tap answers it. The optional line is genuinely optional, and most weeks nobody writes anything — that is fine, because the pattern is what matters, not the prose.
Skipping is not punished and not chased. But three skips in a row is itself a signal, and it reaches a human the same way a low score does.
Not a diagnosis, not a score anyone is ranked on, and never shown to the family a student lives with. A student who fears the answer will be read at dinner will not answer honestly.
Every student can see, at any hour, exactly who is responsible for them — with a face, a name and a way to reach them. “Contact your coordinator” is not support; it is a filing instruction.
Lives near you, has met your host family, and visits in person. Most things end here — which is the point.
A qualified contact, reachable directly — without going through the coordinator if that is the person you would rather not tell.
The organisation configures active staff in order. The product records who was alerted, who acknowledged the request and when the rota ran out; it does not manufacture staffing coverage.
A student can always go around someone. If the problem is the coordinator, or the host family, the path to the mental-health lead and the safeguarding line does not pass through them. A support system that can be blocked by the person causing the harm is not a support system.
Escalation is a server-side clock, not a browser timer. Safety requests start with a 60-second target; after each configured hop the next eligible rota member is alerted until somebody acknowledges the request or the rota is exhausted.
The record is created and the named owner receives emergency in-app, push and email notifications. The server clock starts with that record.
At the default 60-second safety target, the server records the missed target and alerts the next eligible person. The interval can be configured between 30 seconds and 15 minutes.
Each hop names the next active staff member. If nobody remains, administrators receive an explicit rota-exhausted alert telling a person to contact guardians and local emergency services.
Certain categories cannot be closed by one person, cannot be downgraded, and cannot be resolved by the AI at all. Safety and medical always escalate — the option to handle it quietly does not exist in the interface.
Ten months is long enough to change a life. Also long enough to lose someone quietly.
Everything here exists so that the second sentence never happens on our watch.
Wellbeing is the one place where an AI acting on its own would be unforgivable. So on this page, hers is the smallest role in the system — and it is fixed in code, not in a settings panel.
Watch for the shapes a human misses across 1,200 students: three low check-ins in a row, the button pressed twice in a week, a cohort dipping together in week six. She writes one sentence to a named person. That is the whole of her authority here.
She cannot close a wellbeing case, contact a minor's guardian, message the host family about a concern, or decide an outcome. Those tools have no commit function at all — there is nothing for a mis-configured permission to unlock.
And when the AI is switched off, none of this stops. The button, the check-ins, the on-call chain and the escalation clock are ordinary product — they do not depend on a model being available, in credit, or having a good day.
A student will tell you the truth only if they know exactly who reads it. So we show them — on the same screen where they answer, in plain language, every time.
Fields marked as wellbeing or safeguarding never enter an AI prompt, whatever else is switched on, and anything held back is logged as a redaction rather than quietly dropped.
A host family sees what they need to care for someone in their home. They do not see mood history, support requests, or notes about them.
Most placements that fail were already wobbling in week one — jet lag, a language they thought they knew, a house that smells unfamiliar, and nobody to say that out loud to.
Soft-landing camps exist for exactly that week: arrivals tracked flight by flight, rooming that considers who should not be alone, supervision on a named rota, and onward travel confirmed before anybody boards.
By the time a student reaches their host family, they have already met a person from the organisation face to face — so the help button is attached to a face, not an abstraction.
Not “what is your wellbeing methodology”. You are asking: if my child is having the worst night of their life, will somebody pick up — and will I be told?
The request has a named owner, a response target and an automatic escalation trail. Audecius cannot guarantee that an organisation has staffed its rota or that a person will answer; immediate danger belongs with local emergency services.
Safety and medical requests cannot be resolved by the AI and follow the organisation's safeguarding process. Guardian contact depends on that process and applicable duties; the product records the action instead of claiming it happened automatically.
Every incident has a category, a severity, an owner and a resolution. You can ask what happened six months later and get an answer, not a memory.
Audecius Exchange is not a clinical service. We are not therapists, we do not diagnose, and we do not treat. What we build is the infrastructure around the humans who do: routing, timing, records, visibility and accountability.
We do not claim to detect a crisis. Pattern signals are class C0 — they are hints for a person to follow up, and they will sometimes be wrong in both directions. A quiet student is not necessarily a struggling one, and a struggling student does not always go quiet.
Displayed times are software escalation targets, not guaranteed human response times and not emergency-service SLAs. In a medical or police emergency, local emergency numbers come first.