MedUI is open-source software a general practice hosts on its own systems, beside Best Practice. It presents your own record more usefully and can prepare AI drafts for a clinician to review. Before anything goes to a web-based AI model, MedUI removes the identifiers — or use a local model, and nothing leaves the practice. You choose the prompt, the provider and the model.
Redaction is an aid to your judgement — not a guarantee that every identifier is caught.
What it is
Practices are already reaching for AI tools. MedUI is installed on the practice’s own server and reads Best Practice through a read-only connection. It presents the practice’s data to its own staff, and can prepare drafts with a large language model for a clinician to read, edit and decide on. Before a request goes to a web-based model, MedUI removes the patient’s identifiers, the names of the practice’s clinicians and staff, and other people named in the record — and you can see the prompt and context it sends.
The source will be published in full, so you can read the redaction rules, see what they look for and check the behaviour yourself. Names, dates of birth, addresses, phone numbers, Medicare and hospital record numbers are replaced with clear placeholders.
Point MedUI at a web-based provider, or run a model on your own hardware, where nothing leaves the building and redaction is optional. Prompts, providers and models are yours to set, swap and change.
Automated detection is heuristic. It will miss things, especially in free text and in unusual formats — so reviewing the output remains your call. MedUI helps you apply your own judgement; it does not promise that nothing slips through.
Redaction is a deliberate risk-reduction step, not a guarantee of anonymity. MedUI redacts text on your own server, before it is sent to a web-based model, and shows you what it will send — so you can check what it removed and look for anything it missed. (Audio sent to a cloud transcription service can’t be redacted; local transcription is supported.)
Hosted by the practice, alongside a local model if you like, so nothing leaves your network at all.
Read the redaction rules, audit the behaviour, and add your own identifier labels and lists.
Every AI output is a draft for a clinician to review. Nothing is filed, sent or acted on automatically.
No tracking pixels, no session recording, no advertising SDKs.
Automated de-identification is heuristic and incomplete by nature. Identifiers can be missed in free text, in unusual formats, or where information is identifying only in combination. Review the redacted output before you send it anywhere, and never treat MedUI as your only safeguard.
MedUI will be published as open source, so you can read exactly what it does and judge its limits for yourself. Terms of use will be published at launch.
Expression of interest
MedUI is in development, with public release planned for December 2026. Leave your details and we’ll be in touch with early access to the source, setup help, and a walkthrough before release.
Prefer email? Write to interest@medui.com.au.
We’ve recorded your interest and will be in touch with early access details.
About
MedUI began with a simple problem: the AI tools worth using ask for the whole clinical picture, and the whole clinical picture is not ours to hand over. So a general practitioner built the software the practice needed — one that runs on the practice’s own hardware, works from its own record, takes whatever model it is pointed at, and shows what it removed before anything leaves. It has been used in the developer’s own practice since January 2026 and is not yet supplied to anyone else.
It will be published as open source, free of charge, because software that sits between a practice and its patients’ records should not be a black box. The source repository, documentation and setup instructions will be linked here at launch.