Vision & philosophy

Clinical AI belongs at the bedside.

Trustworthy AI that runs where care happens — beginning with AI-native imaging, growing into an agentic, and ultimately embodied, clinical assistant.

public/images/vision.jpg

Photoreal, ~16:11 landscape. A clinician at a workstation in a clinical reading area, calmly reviewing on-screen findings, attentive posture, the screen content abstract and unreadable. Cool clean light with subtle green (#00BF63) accents on the monitor edge. Composed, trustworthy, point-of-care mood. No readable screen text, no patient-identifiable data, no real medical images, no third-party or competitor logos or UI, consented or faceless model.

Why on-site AI

AI should run where the patient is. Keeping inference on-site is a safety, privacy and trust decision: sensitive data stays where it belongs, and clinicians stay in control of a tool they can audit.

Our trajectory

We begin with AI-native imaging — assistance at the point of care. That foundation grows into an agentic clinical assistant that works across the clinical workflow, and ultimately toward an embodied, physical assistant. Each step is grounded in what came before.

Principles

Deterministic and auditable by default. Assistive, never autonomous — the clinician decides. Local-first, so patient data stays on-site. Accessible, so the tools work for everyone.

What we believe

Care happens in a room, not in a data centre.

Medicine has always been a face-to-face craft. A clinician looks, listens, weighs uncertainty, and acts — usually in minutes, often with incomplete information, and always with a person in front of them. For all the progress in computation, that fundamental scene has not changed: the decisions that matter most are still made at the bedside, in the procedure room, or at the microscope. We think AI should meet clinicians there, on their terms, rather than asking them to send their work somewhere else and wait for an answer to come back.

That conviction is why everything we build is local-first. When analysis runs on-site, the tool becomes part of the room rather than a service somewhere on the internet. It is faster, because nothing has to make a round trip. It is more resilient, because care does not stop when a connection drops. And it is more private, because sensitive imaging and patient information never have to leave the building to be useful. Privacy, in this view, is not a policy bolted on at the end — it is a consequence of where the work is done.

It would be easy to treat these as engineering preferences. We treat them as clinical commitments. A tool a clinician cannot trust is a tool a clinician will quietly stop using, no matter how impressive it looks in a demonstration. So we hold ourselves to a simple test: would a careful, accountable clinician be comfortable relying on this, and able to explain afterwards exactly what it did and why? If the answer is anything short of yes, it is not ready.

public/images/vision-bedside.jpg

Photoreal, ~16:11 landscape. A clinician and a patient together in a calm treatment room, face-to-face interaction, equipment present but not dominant, the room clearly the setting where decisions are made. Cool clean daylight, subtle green (#00BF63) accents on equipment trim. Human, present, grounded mood. No patient-identifiable data, no readable screen or chart text, no third-party or competitor logos or UI, consented or faceless models.

Our trajectory

From imaging, to agentic, to embodied.

We start with AI-native imaging at the point of care, grow into an agentic clinical assistant across the workflow, and build toward an embodied assistant at the bedside — each step grounded in the one before it.

A direction we are building toward, step by grounded step.

public/images/vision-trajectory.jpg

Photoreal, ~16:11 landscape. A clean, modern clinical-research space that hints at progression: imaging workstations in the foreground giving way to a sense of advanced robotics in the soft-focus background, no specific robot brand or recognisable product. Cool clean light with subtle green (#00BF63) accents. Forward-looking, careful, optimistic mood. No patient-identifiable data, no readable screen text, no third-party or competitor logos or UI, no people or faceless models.

Where this is going

We earn each step before we take the next.

An honest roadmap is one you can hold the people behind it to. Here is ours, in plain language.

We begin with imaging because imaging is where assistance can be most concrete and most checkable. A finding is highlighted, a measurement is structured, a draft is prepared — and a clinician confirms or corrects it on the spot. There is no mystery and no leap of faith. This is the stage where a tool proves, case after case, that it makes the work a little faster and a little clearer without ever taking the decision out of human hands.

From there, the natural next step is breadth. A single moment of insight is useful; an assistant that follows the thread across an entire workflow is transformative. The agentic stage is about exactly that — connecting capture, review, and record so that a clinician spends less time stitching tools together and more time on judgement. Crucially, this happens under supervision. The assistant proposes the next action; the clinician decides whether to take it. Nothing acts on its own.

The agentic stage has a name: Myro, the clinical assistant that listens, sees and reasons across the workflow. And beyond it lies the bedside. An embodied assistant — present in the room, able to perceive and to help in the physical world — is the most ambitious thing we describe, and we describe it with care. We picture Myro as the brain of a future healthcare humanoid robot: physical AI at the point of care. It is a destination, not a promise for next quarter, and it is meaningless without the trust built at every stage before it. We would rather under-claim and over-deliver than the reverse. Each step is grounded in the one before it, and the clinician remains in control at every single one.

Capability compounds: each stage rests on trust earned in the one before it.

Principles

Auditable

Deterministic and reproducible by default, so every result can be traced.

Assistive

Never autonomous — findings support the clinician, who makes every decision.

Local-first

Inference runs where the patient is, so sensitive data stays on-site.

Accessible

Designed so the tools work for everyone who delivers care.

Assistive by design

The AI proposes. The clinician decides.

Our tools surface findings and structure the busywork, but they never act on their own. The clinician reviews, approves or overrides — and stays firmly in the loop at every step.

Findings are proposed, never imposed.
The clinician approves or overrides.
Every step is logged and reproducible.

Who this is for

Good clinical tools should not be a luxury.

The benefits of clinical AI tend to flow first to the best-resourced places — the large centres with the newest equipment and the deepest benches. We think that is exactly backwards. The clinician working a long list with limited support, the smaller practice without a research department, the team carrying more than its share of the load — these are the people for whom a careful, dependable second set of eyes can matter most. Designing for them is not charity; it is a design constraint that makes the tools better for everyone.

Accessibility also means being honest about what the tools are and are not. We describe intended use plainly, we keep outputs reproducible so they can be examined, and we resist the temptation to dress up support as something more. A clinician should never be left guessing whether a result is a suggestion or a verdict. It is always a suggestion. The verdict is theirs.

Bring clinical AI on-site.

Explore how on-site, assistive AI works across endoscopy, pathology and radiology.

Explore the platform Myro