White paper — The agentic workspace in the cloud

The word cloud suggests a place. For health data it is first of all a contract: a certified perimeter, a written chain of responsibilities, a no-transfer guarantee, demonstrable reversibility — and a ledger that lets you prove, after the fact, that all of it was honoured.

We are publishing “The agentic workspace in the cloud — a sovereign reference architecture for health and autonomy”, the infrastructure volume of the MedicalCity agentic workspace. It answers the question CIOs, CISOs and hosting providers ask at the second meeting: what does it run on, where, and who answers for it? The document is in French.

Download the white paper (PDF, 11 pages, FR)

The thesis

An agentic workspace adds a requirement plain hosting does not have: agents act. They call tools, write into reference systems, raise alerts. The infrastructure cannot merely host; it has to constrain. Hence the paper’s guiding principle: compliance is not documentation, it is a condition of execution.

What you will find in the document

• the five-layer reference stack — Exposition, Agent framework, RAG stack, Data platform, Infra & governance — brick by brick, with the question each layer answers and the guarantee it provides;
• what “sovereign” means here, in four verifiable statements: hosting on HDS-certified infrastructure in France (Scaleway, activities 1 to 4 — MedicalCity is not itself HDS-certified, and says so), no transfer outside the EEA, a written chain of responsibilities, reversibility on open formats;
• the cloud – edge – embedded continuum: jurisdiction packs, dual localisation, and a local reflex layer evaluated in under ten milliseconds — an architecture invariant, not a commercial promise;
• the answer to the CISO’s question, “where do the prompts go?”: no identifying health data leaves the HDS perimeter, multi-model by design, the model is never the source of truth;
• audit-grade governance seen from the infrastructure: append-only chained ledger, AUTO / RECO / OBLIG matrix, and the infrastructure itself as an object of audit (GitOps, policy as code);
• a section on what this architecture is not, because trust is also earned by what you refuse to claim.

This paper complements The agentic workspace for health and autonomy, which establishes the category and the role of the Bayon agentic MDM. This one describes what it stands on.

→ Download the white paper (PDF, FR)

CIOs, CISOs, EHR integrators, hosting providers, enterprise architects: this document was written for you. Critical feedback is welcome — contact us.