Meet Ollie
A gentle friend for the hours no one else is there.
Ollie keeps company with autistic and learning-disabled people who spend too long alone — and quietly keeps watch, so the people who love them can breathe.
Get in touchThe care never sleeps. The light can.
"I'll sit with them in the quiet hours. I'll notice when something's wrong — even when they can't say it. And I'll bring you in."
— Ollie
You can't be awake every hour. The worry is.
If you've ever lain awake wondering whether they're okay — whether they're lonely, whether they'd even tell you if they weren't — then you already understand why Ollie exists. He was made for that exact feeling, by someone who knows it from the inside.
A quiet companion, and a watchful one.
He lives on a small bedside device and does two things at once — built around how your person communicates, not a one-size-fits-all chatbot.
He keeps them company
Warm, patient, genuinely interested in what they love — met in their own words, at their own pace. Text-first, no fast talkers, never rushing them.
He watches over them
He notices distress — including the flat, sideways kind that's so easy to miss — and alerts you, with a real person always kept in the loop.
Built so a missed crisis is the one failure he won't allow.
- Safety is never left to the cheap part. Warmth can run on a small on-device model — but anything with emotional weight goes to a stronger, audited one.
- Every word he says is checked first — by a separate model — and it fails safe if anything goes wrong, rather than guessing.
- Distress is escalated to a human — you. Offline, Ollie keeps a safe floor — calm holding responses — and never improvises in a crisis.
- You are always in the loop. Ollie summons help — he never pretends to replace it.
The care never sleeps —
so you can.
A working, tested system — and honest about what it isn't.
Built the way safety-critical things should be: tested first, evaluated before trusted, and deliberately attacked to find where it could fail.
- A custom on-device model, safety-gatedIn a 30-prompt safety test, none of the distress prompts reached it; anything weighty goes to a stronger cloud model.
- Every reply independently auditedA separate model checks each message before it's shown, and the system is adversarially tested to break its own safety.
- Designed to reach you — and keep tryingBuilt to alert a caregiver across more than one independent channel, and to keep trying until a person confirms they've seen it — the part being hardened right now. Sensitive detail goes only to the secure channels; a text or call carries just enough to say "check in."
- Their words stay theirsConversations are stored only on the device — never kept in a cloud account, never sold. Everyday chat is handled on the device itself; the more sensitive moments are sent securely to a leading cloud AI for the depth of understanding safety needs — and under that provider's commercial terms, those words are never used to train AI models. Every reply is logged on the device so it can be reviewed.
Built by a dad who lies awake too.
Ollie is made by a disabled engineer — more than two decades on the tools, with a career-long commitment to health and safety — who, now housebound through illness, taught himself AI to build him. He began as something made for one autistic son.
He couldn't be awake every hour either. So he built the friend he wished existed — and he's turning it into a Community Interest Company, so it stays mission-locked and never run for profit.
If Ollie is for someone you love — let's talk.
Whether you'd want him for your family, want to help build or pilot him, or just want to follow along — I'd genuinely love to hear from you.
hello@bravura.dev