The short version
Emma is unusual for an AI assistant in three ways. Her intelligence runs entirely on your phone. She answers from a reviewed library of trusted health guidance rather than the open internet. And she shows you the source behind her answers — and tells you honestly when a question falls outside what she can safely answer, instead of guessing.
This page explains how that works, in plain language. None of it requires you to understand the technology — but if a clinician, a family member, or a curious mind wants to know what's going on under the hood, here it is.
1. Everything runs on your phone
Most AI apps send your questions to a company's servers, where a large model produces an answer and sends it back. Emma doesn't. Emma's AI models are downloaded once when you first set her up, and from then on they live on your device.
When you ask Emma something, the answer is produced on the phone in your hand. Nothing is sent to a server to get a reply — which is why Emma works with no signal at all, on a train or a plane or in a clinic with no reception. It's also the heart of her privacy promise: your questions and your health record never have to leave your phone. You can read the detail in our Privacy Policy.
2. A reviewed library, not the open web
Emma does not browse the internet, and she is not a general chatbot that will answer anything about anything. She answers from a curated knowledge base built into the app — a library of eye-health and glaucoma guidance written in plain English.
That library is drawn from government and public-domain health sources only — guidance we are free to reuse and attribute, so you always know where an answer has come from. It includes:
- the UK's NHS, under the Open Government Licence;
- MedlinePlus, from the US National Library of Medicine (public domain);
- the National Eye Institute, part of the US National Institutes of Health (public domain);
- the US Food & Drug Administration (FDA), for plain-language medicines information (public domain);
- and other openly-licensed government health sources, such as the US CDC and Health New Zealand.
We deliberately use only these licence-clean sources rather than sweeping in everything on the web. A smaller, reviewed set is far easier to keep accurate and trustworthy — and it means Emma never relies on material we don't have the right to use. (That's also why you won't see content here from professional societies or charities: we'd need their permission, and we'd rather build on sources that are openly reusable.)
3. How answers stay grounded
When you ask a question, Emma searches her library for the passages most relevant to what you asked, and builds her answer from those — then shows you the source behind it, so you can read the original yourself.
We choose our words carefully here. Emma is designed to ground her answers in her library and to show you where they came from — but no AI is perfect, and Emma is a companion, not an oracle. The source is shown so you can check it, and important decisions about your care should always be confirmed with your own eye-care team.
4. When Emma doesn't know
Some questions sit outside what Emma can responsibly answer — because they need a diagnosis, a personalised treatment decision, or simply because her library doesn't cover them well enough.
5. A safety boundary built in code
Emma's limits aren't left to the AI's judgement.
Emma is designed never to give personalised medical advice, and never to interpret your own readings ("your pressure of 28 is high" is exactly the kind of thing she will not say). That boundary is enforced in the app's own logic — a deterministic safety layer around the AI — not left to the model to remember on its own.
And for anything that could be an emergency — a sudden change in your vision, severe eye pain, halos around lights, or nausea with a red eye — Emma's job is to tell you to seek urgent care now (call 999 or 111 in the UK, or go to your nearest eye casualty), not to calmly explain it away.
6. What Emma is — and isn't
- Emma is an intelligent assistant — software — not a person, despite her friendly name and face.
- Emma is a companion, not a doctor. She explains, reminds and reassures; she does not diagnose, prescribe, or replace your clinician.
- Emma is not a medical device. She does not diagnose, treat, or give personalised medical advice. See our Terms & Medical Disclaimer for the full statement.
7. Which phones Emma needs
Because Emma's intelligence runs on the phone itself rather than in the cloud, she needs a reasonably capable, reasonably recent iPhone with enough memory to run her models comfortably.
We'll publish the exact supported-device list when Emma launches, and the App Store will only offer Emma to you if your iPhone is supported — so there are no surprises after you install her. (An Android version is something we're exploring for the future, but it isn't part of this launch.)
8. How Emma improves
Emma gets better over time through ordinary app updates from the App Store — clearer answers, new guidance in her library, and new features. When her library is updated, the new material is reviewed before it ships.
Emma is a one-time purchase, and those updates are included for the life of the app — there is no subscription. To be precise about a phrase you might see elsewhere: when we say Emma "improves," we mean we improve her in updates. She does not train on you or learn from your conversations — your questions never leave your phone, so there's nothing for her to learn from in the first place.