Clinical data under GDPR and MDR
Strict constraints on residency, audit trail, clinical validation, CE marking.
Healthcare
On-premise clinical AI, MDR, sovereign data.
Hospitals, health authorities, research hospitals (IRCCS), private clinics and biomedical research centres that want to use AI on clinical data without letting it leave the perimeter. noze comes from years of projects with Meyer, CNR and European research centres. Public bodies with no clinical activity will find their page under Public Administration.
Your challenges
Fragmented records
Reports, discharge letters, imaging, PDFs: data scattered and hard to query.
Decisions you can defend in an audit
AI has to support the clinician, cite its sources every time and leave a verifiable trail for every answer.
Post-discharge follow-up
Continuity of care hard to manage without dedicated tools.
How we can help
Product
Product
Product
Product
Solution
Solution
ProductAgenticHealth
Local clinical chatbot, RAG on FHIR/DICOM, digital twin, SaMD MDR pathway.
ProductDataGovern
GDPR and MDR on clinical data: gap analysis, AI risk classification and audit trail, on-premise.
ProductCyberAgent
Vulnerability assessment and pentest of the hospital network and medical devices.
ProductAdmina Enterprise
Clinical AI governance with Admina Enterprise: AI Act, NIS2 and sovereign health data.
SolutionDigital Health
Consulting on clinical roadmap, validation and CE pathway.
SolutionResearch & Development
25+ years of applied biomedical research, European projects.
Related projects
2019–2023
ALA (Advanced Laboratory Automation)
Joint R&D INPECO + BioRobotics Sant'Anna Pisa (EUR 3.5M, 30 researchers): automation and robotics for clinical labs, deep learning on skin lesions.
2008–2010ESCOGITARE: Collaborative Bioinformatics
noze system for ISTI-CNR: bioinformatics workflows on an existing computational grid, drag-and-drop Visual Workflow Designer and Web 2.0 CMS.
Insights
Agent harness and sandbox: TrueForge and smolvm
TrueForge is an open-source harness in TypeScript holding the execution loop, MCP tools, approvals and secrets. smolvm is a Rust CLI booting microVMs with a separate guest kernel on libkrun, with an egress allowlist in the configuration file. Underneath sit five different isolation levels, from a syscall filter to a VM, and each stops different things.
GLM-5.3: new post-training and cyber benchmarks
Z.ai announced GLM-5.3 on 14 August, stating it starts from the same base model as GLM-5.2 and that every gain comes from post-training. The weights are not out yet. On CyberGym and ExploitGym the benchmarks measure something other than finding a vulnerability, and the CyberGym paper reports 759 raw crashes for 9 confirmed zero-days after manual triage.
Zero-knowledge proofs over inference: the Attestable launch and the published numbers
Attestable announces a $20M seed and writes that practical zero-knowledge proofs were considered impossible and that the company solved it. The technical post declares two temporary limitations, a 16K token window and integer quantisation of matrix multiplications, and labels the benchmarks alpha results on a single H100. IACR ePrint 2025/535, a preprint with no experimental setup described and no peer review, states 150 seconds per token on one core for Llama-3 8B.
Frequently asked questions
Can you use AI on clinical data without letting it leave the hospital?
Yes. agentichealth is a local clinical chatbot with RAG on FHIR and DICOM: data stays inside the organisation's perimeter, on-premise. datagovern adds GDPR and MDR gap analysis, AI risk classification and an audit trail, all on-premise.
Does AI-based medical software need CE marking?
If it qualifies as a medical device, yes, and it has to follow the SaMD pathway under MDR. agentichealth is designed along the SaMD MDR pathway, and the healthcare consulting covers clinical roadmap, validation and the CE pathway.
How do you make a clinical AI answer defensible in an audit?
You need AI that supports the clinician, cites its sources every time and leaves a verifiable trail for every answer. Admina Enterprise brings clinical AI governance across the AI Act, NIS2 and sovereign health data; datagovern keeps the audit trail on-premise.