Open seed · MIT Published, not hosted Rev 0 · specification

AORTA · the other half

KINDRED

Fifty-five organizations exist because somebody said yes on the worst day of their life. This is the half that is about them. Volunteer peer support for donor families, transplant recipients, and the people still waiting — from someone who has actually been there.

§ 01

Three groups, and they are not one group.

Everything else on this site is about the organization — the coordinator, the record, the board, the case under a clock. This is about the people the case was always for, and they arrive in three very different states.

Donor families

Someone said yes in the worst hour of their life, and then everyone went home. The support usually stops long before the grief does.

Recipients

Living on immunosuppressants, carrying a gift that came from a loss, often with a complicated gratitude nobody around them quite knows how to hear.

Still waiting

On a list, for an unknown length of time, for a call that may not come. A distinct kind of waiting that people who have not done it cannot really picture.

A bereaved parent and a waiting recipient need different rooms, different tone, and sometimes protection from each other's timing. Treating them as one audience is the first mistake, and it is the one almost everybody makes.

§ 02

Someone who has actually been there, at three in the morning.

A member asks for support and is matched to a trained volunteer who shares their situation — matched on organ, on where they are in the journey, on region. Then they talk, one to one, over video.

Not a forum. Not a comment thread. Not a chatbot. A person who has been through the same thing, available when it is actually needed — which is rarely during office hours.

The value here is not the software. It is that the person on the other end has buried someone, or carries a transplanted organ, or is waiting too.

Around that call sits the rest of it: vetted educational material, a library of recorded conversations, and — quietly, in the background — the machinery that makes a 3 a.m. match actually happen.

§ 03

Agents run the building. Humans do the meeting.

It is 2026, and the obvious question is why the volunteer has to be a person at all. The answer is specific rather than sentimental.

In peer support, the human is not the delivery mechanism. The human is the product. The thing that works is shared lived experience — identification, one survivor to another. An AI has not had a transplant, has not buried a child, has not waited on a list. An AI "volunteer" is not a cheaper version of this; it is a different thing wearing the same clothes, and in a moment this raw, someone who later realises they were consoled by a machine that implied it understood has been harmed by us.

No agent may claim, imply or perform lived experience. Ever. And every agent says what it is, unprompted, every time.

That rail is narrow on purpose — because outside it, the case for AI here is strong:

What agents do — nearly everything

  • The dispatcher. Nobody is awake at 3 a.m. to match a request, notice a volunteer never joined, or chase a stall. This is what makes 24/7 true instead of aspirational.
  • Triage. Is this an anniversary, or a crisis? Route accordingly.
  • A safety watch that surfaces to a human moderator — and can never intervene in the call itself.
  • Answering questions that are information rather than support, with citations — which frees the volunteers for what only they can do.
  • Volunteer care. Noticing burnout before the volunteer does.

What only people do

  • The conversation itself.
  • Deciding that a volunteer is ready.
  • Answering an escalation.
  • Anything that changes a person's standing.
  • Sitting with somebody at 4 a.m. and saying I know, because it happened to me too.

And when nobody is free at 4 a.m., the honest thing is not a bot pretending to be a peer. It is an agent that says plainly: no one is available right now — I can stay with you while you wait, I can page the on-call, or here are crisis resources. Presence without pretence.

§ 04

What it never does.

These are structural, not settings. Several are enforced by the shape of the database rather than by a policy somebody could relax under deadline.

KINDRED never

  • Records or transcribes a call. The ledger holds that a conversation happened — never what was said. A member who thinks they might be recorded will not say the true thing, and the true thing is the entire point.
  • Lets an agent pose as a peer. §03.
  • Optimises for engagement. No streaks, no nudges, no daily-active anything. Success here is somebody needing this less.
  • Puts a donor family and a recipient in touch directly. That contact is mediated by the OPOs, for good reasons — so the permitted pairings are enumerated in the schema and the forbidden one has no row to select.
  • Invents a crisis number. Or an emergency contact, or a clinical fact. The system refuses to start with those unset rather than starting with a plausible placeholder.
  • Lets anyone make themselves a volunteer. Standing is granted after a human signs, server-side, and the database refuses an unsigned volunteer.

§ 05

The hard part isn't the code.

In 2026 an app like this is a few weekends. This is the part that isn't, and it is the reason the project is given away rather than run.

It puts grieving, medically fragile strangers into private video calls. Both sides are vulnerable — the volunteer is a survivor of the same thing, which is what makes it work and also what makes them fragile.

Vetting is a pipeline, not a checkbox

Identity, background check, training, assessment against hard rehearsal scenarios, and a named human's signature. Plus one rule people forget: a volunteer must be a defined distance from their own loss before supporting anyone else. Someone three weeks out is not ready, however much they want to help — and wanting to help is often the symptom.

Crisis will happen

Someone will disclose suicidal ideation, or abuse, or a medical emergency, to a volunteer who is not a clinician. There must be a staffed escalation path with a person at the end of it, a script, and a debrief afterwards — because someone who has just taken a crisis call is now a person who needs support. If the service claims 24/7, the escalation path is 24/7. If it cannot be, it does not claim it.

Before a single real member is admitted

A legal entity exists and holds the app store accounts.
Insurance appropriate to the activity is in place.
A clinical or bereavement-services advisor is named and reachable.
The crisis escalation path is staffed for every hour the service claims.
Crisis resources are supplied and verified for every jurisdiction served.
Volunteer vetting runs end to end, with a named human signing.
Moderation is live — block, report, a human queue, published contact.
Deletion works, tested against a real account.

THE FULL CHECKLIST IS TWELVE ITEMS AND IT IS IN THE REPOSITORY. THEY ARE QUESTIONS ABOUT AN ORGANISATION, NOT ABOUT SOFTWARE — WHICH IS EXACTLY WHY THEY MATTER MORE THAN THE CODE DOES.

§ 06

Published rather than hosted — and that is the point.

Running this means being the person responsible when a call goes wrong at two in the morning. That duty is organisational, not technical: it needs an entity, insurance, a clinical advisor, a staffed phone. No amount of good code discharges it.

So this is not a service, and there is nothing to sign up for. It is a seed — a specification complete enough that a team can point a coding harness at it and begin, carrying the parts a harness cannot work out on its own: the domain constraints, the safeguarding spine, and the laws that make it safe to build at all.

Open-sourcing this is not stepping back from the duty of care. It is routing it to somebody who can actually carry it.

If that is you — a transplant centre, an OPO's family-services team, a patient organisation, a foundation, or a developer who has been through this themselves — take it. Fork it, rename it, ship it as your own. The licence permits it and the author will not be involved.

Two asks, neither enforceable, both meant: satisfy the readiness checklist before you open to real members, and if you find something here wrong, say so in public where the next organisation can read it.

STATUS: REV 0. A SPECIFICATION, NOT AN APPLICATION — THERE IS NO CODE YET AND NOTHING HAS BEEN RUN WITH REAL MEMBERS. THE DESIGN WORK USED THE WORKING NAME LEGACYCONNECT; A SEARCH FINDS A REGISTERED TRADEMARK OVER MOBILE-APP SOFTWARE UNDER THAT NAME, SO WHOEVER CARRIES THIS SHOULD CHOOSE AND CLEAR THEIR OWN.

Everything else here is what the organization does.

This is who it was for.

The record gets kept, the work gets watched, the conversations get rehearsed — and then a family goes home to a house with someone missing from it, and a recipient wakes up carrying a gift they can never thank anyone for, and somebody else is still waiting for the phone to ring.

That part has never had software built for it properly, and it is not because the software is hard.