Data Intelligence. Deliberately narrow.
A compliant archive is a cost centre until you can find anything in it. This layer makes held data searchable, classifiable and measurable — and stops firmly short of clinical decision support, because that is a regulated medical device and we do not pretend otherwise.
- Status
- Partial GA
- Inference location
- Switzerland
- Trained on your data
- Never
- Default state
- Off
Four capabilities ship today.
Six are on the roadmap, dated.
Healthcare software is sold with too much future tense. On this page, anything marked Available is running in production for paying customers today. Anything marked with a quarter is not built yet, and we will not invoice you for it. If a capability slips, we update this date rather than quietly removing it.
A compliant archive is
a cost centre until you can find anything in it.
Four capabilities for finding, monitoring and classifying what you already hold are live today. The rest is dated roadmap, not a demo dressed up as a product.
- 4
- Capabilities available now
- 6
- Roadmap items, dated
- Off
- Default state
- Never
- Trained on customer data
Running in production
for paying customers.
Metadata & record search
Search across patient identifiers, encounters, document types, dates, authors and coded fields, always filtered by what the searcher is authorised to see.
Access anomaly detection
Behavioural baselines per role, flagging unusual volume, timing, breadth or export patterns. Live today in the Audit platform.
Operational dashboards
Record growth, access volume, integration throughput, error rates and storage consumption, per site and per department.
Data quality monitoring
Missing identifiers, duplicate patient candidates, orphaned documents, coding gaps and stale records surfaced as a work queue.
Not built yet.
Dated anyway.
These are committed roadmap items with target quarters, not aspirations. Enterprise customers can influence sequencing through the quarterly service review.
Federated full-text search
Search inside document content across the vault, with results filtered per-user by the same policy engine that governs record access.
Automatic document classification
Classify incoming documents by type, specialty and urgency on ingest, so filing stops being a manual administrative task.
Medical document indexing
Structured extraction of key fields from unstructured clinical documents, with confidence scores and mandatory human confirmation.
Administrative summarisation
Summaries of correspondence, referral threads and administrative history — explicitly not clinical summarisation.
Workflow automation
Rule- and event-driven routing of documents, referrals and review tasks between departments and external providers.
Compliance intelligence
Predictive flagging of governance risk: policies drifting from practice, review backlogs, and access patterns trending toward a finding.
Where we stop,
and why.
The most important part of an AI capability in healthcare is the line it refuses to cross. Ours is written down, contractual and auditable.
- What this layer does
- Administrative and operational intelligence: finding things, classifying things, routing things, and spotting operational or governance anomalies.
- What it explicitly does not do
- Clinical decision support. No diagnosis, no triage, no treatment recommendation, no risk scoring of patients, and no output presented to a clinician as clinical guidance.
- Why we draw the line there
- Clinical decision support is a regulated medical device under Swissmedic and the EU MDR. It demands clinical governance, evidence and post-market surveillance we have not built. We will not ship it informally.
- Human confirmation
- Every extraction and classification carries a confidence score and enters a review queue. Automated output never silently becomes part of a clinical record.
- Training data
- Customer patient data is never used to train models — not for your tenant, not for anyone else's, not in aggregate, not de-identified. This is a contractual commitment in the DPA.
- Processing location
- All inference runs inside Swiss infrastructure. No patient data is sent to an external model provider, in any jurisdiction, for any purpose.
- Zero-knowledge tenants
- Content-based intelligence cannot operate on data we cannot read. Zero-knowledge tenants retain metadata search, anomaly detection and operational analytics.
- Opt-in
- The Intelligence layer is off by default. Enabling it is an explicit, audited administrative action, revocable at any time.
Shape what
we build next.
If a roadmap item would change your operations materially, tell us. Enterprise customers influence sequencing directly, and we would rather build the thing a hospital actually needs than the thing that demos well.
Intelligence is opt-in, off by default, and never trained on customer data