For hospitals & medical AI

Two things that could never touch. Now touching, safely.

The best model and the patient who needs it never meet: the law won't release the data, the vendor won't release the weights. Noolog seals both in one box in your hospital. Inference happens. Nobody gives up a thing.

Request a pilot node → HIPAA · GDPR · EU MDR · on-prem

The deadlock

The best model and the sickest patient never meet.

Not a tech gap. A standoff — both sides refuse to move first, both for good reason.

Vendor can't ship out The weights are the company.

Years of research, the vendor's whole moat. At the edge it can be copied or extracted. The moment it leaves, control is gone.

Hospital can't ship out The data can't board a plane.

Scans and records bound by HIPAA, GDPR, residency law — and ethics. PHI can't sit in someone else's cloud.

Model in a cloud the data can't reach, data in a building the model can't enter — cross-border medical ML stalls before it starts.

The break

One sealed node ends the standoff — on the hospital's floor.

A tamper-responsive enclosure in the hospital. The model runs inside the HSM boundary; data flows in from the ward. Two crown jewels meet behind one wall neither side can open.

01 · Data Patient data never leaves.

PHI moves from the ward into the node in your building — never a border, never a vendor cloud. Residency and HIPAA/GDPR hold by construction.

02 · Model Weights stay sealed inside.

The model runs inside the tamper-responsive HSM boundary. Nobody extracts it — not staff, not whoever steals the box. The vendor keeps its IP.

03 · Result Only the answer comes out.

Data in, weights sealed, inference between — only the diagnostic result crosses the wall, trailing an append-only audit record.

Inside the box

Where the two crown jewels finally touch.

Left: what the hospital could never send out. Right: what the vendor could never send out. They meet inside one sealed boundary; only the result crosses the wall.

HOSPITAL · in the building Patient Data scans · records · PHI ✕ cannot go to the cloud AI VENDOR · off-site Model Weights crown-jewel IP ✕ cannot ship to edge the route that never worked SEALED NODE · IN THE HOSPITAL tamper-responsive enclosure HSM BOUNDARY inference data + weights meet here data in weights sealed in Diagnostic result + audit trail only the answer crosses the wall
Data — stays in the building Model — sealed in the HSM Result — the only thing that leaves

Neither crown jewel crosses a border. Data stops at the wall; weights can't be pulled out. The only export is the result — with a tamper-evident record built for the regulator. How the deliberation works →

Sealed both ways

A closed box is also a clean one.

Because the node is a fully enclosed, sealed device — no vents, no open ports — it runs sterile and dust-free. That opens a direction: eventually a node can sit inside the hospital, right beside the patient, integrated into the medical devices themselves.

The interface, too

The UI is built to the same bar as the box.

Engineered to IEC 62304 (software lifecycle) and ISO 14971 (risk management). That bar is reachable because the frontend is open-source Slint: statically typed, compiles to native Rust/C++, no GC runtime, deterministic, with a GUI test-automation framework. Not every standard binds every deployment. See the compliance model →

Let the model meet the patient. Finally.

A sealed Noolog node on your floor runs the diagnostics you've been locked out of — no data leaves, no weights leak, the regulator gets the trail by construction.

Request a pilot node → See the compliance model → MDR / FDA SaMD / IEC 62304