How it's set up
The four lines under the hero, with their evidence. Everything below is a property of the running system, not a statement of intent. Ask us to show you any of it on a call.
Last updated 9 August 2026
Where your data is held
The application, your records and the AI that reads them all run inside the EU.
The part worth asking about, and the part most vendors gloss over, is the AI. Sending a European record out of Europe to be interpreted has not kept anything in Europe, whatever the hosting page says. Ours does not leave.
Encryption
Everything is encrypted on the way in and out, and everything is encrypted where it is stored. Keys are managed by the platform rather than by hand, and there is no unencrypted path through the system at all.
Who can sign in
Accounts require two-factor authentication and a verified email address. Practices that want a harder edge can put the whole application behind a second gate that checks identity before anything loads, so a request from someone we cannot identify never reaches the application at all.
Sessions are cached with a deliberately short lifetime, so a revoked account loses access in a window measured in minutes rather than at the next natural expiry.
Who can see what
Every record is scoped to a practice. Within one, access is governed by role (owner, admin, member) and layered with the clinical roles a practice actually uses: doctor, nurse, receptionist, assistant, administrator. People join by invitation, and an invitation can be withdrawn before or after it is accepted.
Files
Nothing in storage is publicly addressable. Every image and document is delivered through a signed URL that expires, so a link copied out of the application stops working rather than becoming a permanent door into the record.
The same applies in reverse to the patient upload links. A link or QR code you generate is scoped to one purpose (a profile photo, clinical media, or documents) and carries an expiry, a size ceiling, an allowed file-type list and a limit on attempts. It grants nothing beyond the single upload it was made for.
Dictated audio
Two options, chosen per user in settings. The default transcribes in the same European region as everything else. The alternative keeps the recording inside your own system. It is turned into text by the same service that runs the application, and never reaches an outside transcription provider at all.
If dictated audio is the sticking point in your assessment, this is the setting that resolves it.
Application controls
- Anything that changes data is checked for where it came from, so a request forged on another site is refused.
- Rate limits on every part of the system, including the AI, so one account cannot exhaust it for everyone.
- Traffic between the parts of the system is filtered rather than passed straight through.
- Individual capabilities can be switched off for one practice, so you can decline a feature without declining the product.
What the assistant is not allowed to do
The clinical limits are enforced by the software, not left to the assistant's judgement. It does not recommend prescriptions. Anything diagnostic is presented as support for your decision, with its reasoning and its gaps shown, and the clinician decides. Emergencies are routed to 112. And it cannot state anything about a patient that it has not actually read from the record, so it cannot invent a history.
Answers drawn from published guidance carry the source and a link back to it, so a claim can be checked rather than trusted.
What we don't claim
We hold no completed HIPAA attestation, SOC 2 report or ISO 27001 certificate today. That is why you will not find those badges under the hero. A certification means a finished audit and a report we can hand you on request, and we would rather tell you this here than have your compliance officer find it out.
The controls above are real and demonstrable. The paperwork that would let us summarise them in a logo is not done. If a formal attestation is a hard requirement for your practice, tell us on the call and we will be straight with you about timing.
Formal assessors get the rest in full, with a named person under an NDA: architecture, subprocessors, data flows and retention.
Reporting something
If you think you have found a vulnerability, email info@medical.ai. Tell us what you found and how to reproduce it. We will confirm receipt within two working days and keep you updated until it is closed. We will not pursue anyone who reports in good faith and gives us a reasonable chance to fix it first.