The professional view of Ollie: what's built, what's in progress, and where we need expert partners.
Ollie is a safety-first AI companion-and-safeguarding system for isolated autistic and learning-disabled adults. This page is written for the people whose job is finding out what's real — because that's exactly what we want you to find.
Initial focus: adults. Any extension to children or young people would be a separate programme, governed by the ICO Children's Code.
A low-cost bedside device, built around a routed safety architecture.
A small on-device model handles everyday companionship. Turns are routed by severity: light, everyday chat stays on the device; anything with emotional weight goes to a stronger cloud model. Every reply is independently audited before it's shown — cloud-audited when online; offline it falls to a curated safe floor only, never a generated guess. Any sign of distress escalates to a named human carer, who is always in the loop.
See the family-facing page, including a step-by-step walkthrough of one message through the system.
Built, drafted, or in formation — stated plainly.
We publish this the same way we track it internally: nothing is called finished until it is.
| Area | Status |
|---|---|
| Technical safety architecture — deterministic crisis guard, routed inference, fail-closed reply audit, adversarial testing | Built and tested. |
| Data protection by design — local-first single-file storage, content-free safety logs, fail-closed signed updates, minimised cloud exposure, no-cloud encrypted backup | Built — verified in the codebase. |
| Provider terms: customer content never used to train models | Confirmed (Anthropic Commercial Terms §B). |
| Data Protection Impact Assessment (UK GDPR Art. 35) | Drafted; professional review and the ICO template still to come. We treat a DPIA as a legal precondition of any real-user pilot, and the draft already exists. |
| Safeguarding Adults Policy — named lead, escalation pathway, Care Act 2014 | Drafted; seeking a safeguarding-experienced reviewer. |
| Clinical sign-off of the pilot design | Not yet obtained — a hard pre-live gate we set ourselves. No real user until a qualified clinician confirms it's safe to proceed. |
| Legal structure | Not yet decided. A community interest company and a charity are both ruled out — neither permits the investment the development needs. We are taking advice on how the remaining options interact with the household’s benefits before registering anything. |
This page exists to find these people.
- A clinician (learning-disability or safeguarding-experienced) to review our structured safeguarding questions and give a focused "safe to proceed" pilot sign-off. We've prepared the exact questions — it's a structured review, not an open-ended favour.
- A safeguarding-experienced charity partner for a co-produced pilot — the charity as accountable applicant with ethics and lived-experience governance, Ollie as the technology.
- A DPO or data-protection reviewer to take the drafted DPIA through the ICO template, including the cloud provider's data processing terms.
- Governance advice — the structure is still open, and we would rather choose it well than quickly. We’d value a steer from anyone who has weighed these options where the founder’s household income is benefits.
- Pilot funding — get in touch and we'll share the current funding pack.
The lines we hold, on purpose.
"Ollie helps detect distress, but does not diagnose or treat."
Ollie is not therapy, and not a crisis service. It does not replace family, carers, the NHS, or emergency services — it raises what it notices to a named carer, who decides what happens next. The AI never makes an autonomous safety decision; a human carer is always in the loop. Ollie is designed and positioned as a companion. The formal software-as-a-medical-device determination is a question we're seeking regulatory input on early, precisely because it shapes everything downstream.
A document pack for partners doing due diligence.
These carry operational and personal detail beyond what belongs on a public page, so they're shared directly rather than posted — email us and we'll send what's relevant to your role.
- Clinical & safeguarding brief, with data-protection information requirements
- Data Protection Impact Assessment (draft)
- Safeguarding Adults Policy (draft)
- Current-build evidence pack