An AI-first clinical workspace.

Clinical software is usually an accretion: a scheduling tool, a records system, a dictation product, a compliance layer — each bolted to the next, with the AI (if any) added last and trusted least. Regulated health data raises the stakes: who consented to what, and who accessed what, can't be an afterthought.

We built the workspace AI-first on a single substrate rather than integrating products. The health record lives on a FHIR-native foundation — the clinical data standard, native rather than translated — and everything the clinic does runs on that one substrate: scheduling, charting, ambient transcription during the visit, and note summarisation after it.

Consent and audit are not features inside the care workflow; they are their own planes. The system separates three trust boundaries — care, consent, audit — so a record crossing from one plane to another passes an explicit checkpoint.

The figure above is the real shape of the build: three lanes over one substrate. The care lane carries the clinical workflow — the scheduling, the chart, the ambient documentation. The consent plane governs what the patient has agreed to. The audit plane records access. The solid line is a record flowing through the system; each diamond is a checkpoint where the record crosses a trust boundary.

The AI work — transcribing the visit ambiently and summarising it into the note — happens inside those boundaries, on the same governed substrate as everything else, not as an external service the record is shipped out to.

One substrate, three trust boundaries. The provider gets a workspace where the AI is structural rather than bolted on — and where consent and accountability are enforced by the architecture, not by policy documents.