Honest by designOne platform · three peopleEU · privacy-first

Never feelalone.

One honest AI companion for people with cognitive differences — and the calm, private tools for everyone who cares for them.

Autism, Down syndrome, brain injury, early dementia. Across every one of these, the same finding repeats: structural loneliness. Folk AI gives the person a warm, unhurried friend that talks like a human — while their guardian and clinician see only mood, topics and safety signals, never the conversation.

Meet the companion →The three experiences

folk.ai/dashboard

GDPR Article 9Handled as special-category data from the first line of schema.
Working toward MDR Class IIaThe certification path we are on. Not certified today.
Safety runs outside the modelA deterministic screen fires first — and fires even if the model is down.
Data stays in the EUEU database, EU hosting, every sub-processor under a signed DPA.

The gap

Between appointments, no one can see how the week actually went.

A specialist sees the person for an hour, sometimes once a month. A guardian sees the hours they are in the room for. Everything in between — the rising anxiety, the quiet withdrawal, the week that slowly got worse — leaves no record. By the next appointment, everyone is working from memory.

01
Asking is harder than it looks
The same communication barriers that make everyday life hard also make it hard to say “I am not okay” in the moment it would matter most.
02
Recall stands in for a record
When the only account of six weeks is what someone remembers, a slow decline looks a lot like an ordinary week.
03
Risk does not wait for the calendar
Self-harm and acute crisis are elevated across these groups, and the moment of need rarely lands on the day of the appointment.

Who it’s for · one shared problem

Five conditions. The same human experience underneath.

It isn’t five products for five labels. Across autism, Down syndrome, acquired brain injury, intellectual disability and early dementia, the same barriers recur — and Folk AI is built around them.

Structural isolation

Connection is harder, rarer and less satisfying than the person wants — not a preference, a structural barrier. A 24/7 companion that never tires or judges.

Systems written to exclude

Letters, benefit forms, tenancy and healthcare documents assume a kind of processing these minds don’t have. Easy-read translation, in the chat, in seconds.

Invisible distress

What happens between rare appointments — rising anxiety, a quiet decline — is invisible today. Wellbeing signals surface it early.

No private space

Most are supervised or supported; private thought is structurally limited. The companion is the one conversation that stays theirs.

Dignity & autonomy

Routinely infantilised and excluded from decisions about their own lives. The companion treats them as the adult they are; consent is real and granular.

Elevated crisis risk

Self-harm and acute crisis are elevated across all these groups, and the same communication barriers make asking for help harder. Three-layer safety net, ~15-20s to a human.

What it is

One product, built around the person — and the people who care.

Folk AI is not a chatbot bolted onto a dashboard. It is a single platform with three connected experiences — the companion the person talks to, the calm window their guardian watches over, and the professional view their specialist relies on — fed by one private engine that turns conversation into signal, and never lets the conversation itself leak.

How it works

Three experiences. One engine underneath.

The person talks. The system derives signal from that conversation. The guardian and the specialist see the signal — each at the depth their role needs, and no deeper.

FOR THE PERSON

The companion

A warm, honest friend that talks like a human, remembers what matters, and stays safe under pressure. Not an assistant — a presence.

Open the companion →

FOR THE GUARDIAN

The guardian panel

Reassurance without surveillance: mood, topics and safety alerts — the shape of the week, never the words behind it.

Open the panel →

FOR THE CLINICIAN

The specialist view

A triage-first caseload and a per-client specialist card — what changed since last session, in objective signal.

Open the clinic →

The RED LINE. A guardian or clinician never reads the conversation — not a summary, not a paraphrase, not an excerpt. Only derived signals. Enforced in Postgres RLS, the API layer, and a verbatim sanitiser. See how →

Why you can trust this

Guarantees live in the system, not in the policy page.

A promise that can be switched off is not a guarantee. These three are constraints in the architecture — remove the marketing and they still hold.

Safety does not depend on the model

A deterministic screen reads every message before any model is involved — it fires instantly, and it fires even when the analytical model is slow or down. A deeper read then grades what it found. A human always decides what happens next.

~15-20s from a high-risk flag to a person

The conversation has no read-path out

No server endpoint returns raw message content. Row-level security in Postgres will not release messages; the API selects aggregates only; an output sanitiser strips any quote the model slips in. Remove one layer, two still hold.

Three independent enforcement layers

Built for the rules it will be held to

Mental-wellbeing data is GDPR Article 9 special category; the lawful basis is the person’s own explicit consent. Under the EU AI Act the product is built as decision support with mandatory human oversight — signal in, a human decides.

MDR Class IIa is the path we are on, not a status we hold

Read the full safety & privacy model →

Explore

Six pages. The whole product, end to end.

Companion →

The chat, persona, crisis protocol, feed, memory, sensory space.

Guardian →

Setup, dashboard, alerts, mood & themes, reports, AI insights.

Specialist →

Triage, client card, the analytical AI that writes DAP notes.

Safety & privacy →

Three safety layers, the RED LINE, GDPR & EU AI Act compliance.

Platform →

The stack, the data flow, and the roadmap.

The model →

Three audiences, unit economics, why the clinic is the door.

In their words

A clinician’s own words belong here.

This space is reserved for one quote from a specialist who has used Folk AI in their own practice.

The first controlled pilot is running with a small number of participants. Until a clinician is ready to put their name to a sentence, this frame stays empty — we would rather show you an empty frame than a quote nobody said.

Questions