AI medical documentation for hospitals. In your cloud, your data centre, or ours.
The same clinically evaluated ambient AI running in German university hospitals, now with HIS integration, on-premise deployment.
Is Nixi AI right for our hospital?
Yes. Nixi AI runs in German university hospitals today and was evaluated in a peer-reviewed EULAR study (documentation time reduced in 82% of 88 consultations). For hospitals and MVZ networks we offer three deployment models: Standard Cloud (BSI C5-attested Frankfurt infrastructure), Dedicated Frankfurt (own tenant with audited open-weight models) and On-Premise (signed Helm chart in your own Kubernetes cluster, air-gapped supported). HIS integration runs via HL7 / FHIR into Orbis, i.s.h.med, CGM Clinical or Meierhofer M-KIS.
Deployment models
Three ways to run Nixi AI safely.
Every hospital has different data-protection, compliance, and IT-governance constraints. We don't offer one model. We offer three, so the operating mode fits the risk profile, not the other way around.
Pick the constraint your data-protection board has set.
Standard (EU cloud)
EU-hosted infrastructure in Frankfurt. Sufficient for most hospitals.
- Data resident exclusively in Frankfurt, no US transfer
- BSI C5-attested cloud infrastructure (mandatory under § 393 SGB V since July 2025)
- § 203 StGB-compliant data-processing agreement, certified external DPO
- Production tenant provisioned in 72 hours
Best for
Hospitals and MVZ networks without an explicit on-premise mandate from internal data-protection boards.
Dedicated Frankfurt
Own tenant, audited open-weight models, traceable audit trail.
- Dedicated tenant in Frankfurt: no shared inference pipeline
- Audited open-weight models with German medical fine-tunes, instead of a closed API
- Customer audit-log sink (S3-compatible, Splunk, Elastic)
- Data-processing architecture diagram and model weights remain auditable
Best for
University hospitals and large networks that must present concrete model cards to data-protection authorities or ethics committees.
On-premise
Entirely inside your Kubernetes cluster. Air-gapped possible.
- Signed Helm chart for Kubernetes 1.27+
- Air-gapped mode with a local model registry: no outbound internet required
- Runs on on-prem GPUs (Nvidia A10 / L4 / H100) or CPU-only for smaller hospitals
- Hybrid mode: sensitive audio stays local, managed UI in Frankfurt
Best for
Hospitals with KRITIS-adjacent exposure, strict data governance, or contractual obligations to keep data inside the cluster.
HIS integration
HL7 and FHIR. Not copy-paste.
The EULAR study was run with manual copy-paste into the chart, and still reduced documentation time in 82% of consultations. Enterprise customers eliminate that last step: Nixi writes the structured note directly back into the leading HIS via FHIR or HL7 v2.
Supported HIS systems
Dedalus Orbis
FHIR R4 + HL7 v2 via NextGen interface
SAP / Cerner i.s.h.med
FHIR R4 + i.s.h.med native adapter
CGM Clinical
FHIR R4 + CGM tenant proxy
Meierhofer M-KIS
FHIR R4 + MEIERHOFER Connect
Other HIS on request
Via REST API with FHIR-compatible resources (Encounter, Observation, Condition, DocumentReference, MedicationRequest)
What is handed to your HIS
Encounter
The visit envelope: patient ID, specialty, timestamps, treating clinician.
Observation
Vitals, DAS28 / CDAI / SDAI scores, structured measurements.
Condition
Diagnoses with ICD-10-GM codes and confirmation status.
DocumentReference
Finalized note as Markdown + PDF + raw text, ready for the chart.
MedicationRequest
Prescription hints (optional, when spoken during the consultation).
Architecture
How Nixi AI is built.
Three logical layers that repeat across all three deployment models. Architecture diagram (Ingress · Storage · Model Host · HIS Bridge) available on request in the RFI pack.
Ingress
Audio capture from browser / iOS / macOS desktop client, TLS 1.3, signed upload URLs, 14-day retention on audio chunks before automatic deletion.
Processing
Transcription workers (open-weight models fine-tuned for German medicine) + clinical NLP workers (structured note generation, ICD-10-GM suggestions, evidence extraction).
HIS bridge
FHIR or HL7 v2 handover to the leading HIS. Audit log events (OIDC identity, encounter ID, action) streamed to the hospital-side SIEM sink.
Departments
Where Nixi AI pays off in the hospital.
Ambient documentation delivers the most where time-per-case is high and note structure is non-trivial.
Rheumatology
DAS28, CDAI, SDAI scores, biologics justifications, triple-therapy histories. Peer-reviewed in the EULAR 2025 study.
Neurology
MS progression notes, epilepsy seizure protocols, NIHSS scores, structured with timestamps.
Orthopaedics / trauma surgery
D-Arzt reports (F1000), OP reports, progression notes. Form-conformant for German workers'-comp billing.
Internal medicine
Discharge letters, echocardiography, and endoscopy findings, ward-round documentation.
Psychiatry / psychotherapy
PTV-3 reports, narrative anamneses, session notes. Narrative structure without fill-in-the-blanks.
Oncology
Tumour-board protocols, treatment plans, complex referral letters, structured with links to pathology and imaging.
Model comparison
Standard vs. Dedicated Frankfurt vs. On-premise.
| Standard | Dedicated Frankfurt | On-premise | |
|---|---|---|---|
| Data residency | Frankfurt (EU cloud) | Frankfurt (dedicated tenant) | Your data centre |
| Model weights | Managed (not disclosed) | Audited open-weight | Audited open-weight |
| BSI C5 basis | ✓ Infrastructure | ✓ Infrastructure + dedicated controls | Your environment |
| Onboarding time | 72 hours | 2–4 weeks | 4–8 weeks + assessment |
| HIS integration | FHIR / HL7 v2 | FHIR / HL7 v2 + custom | FHIR / HL7 v2 entirely inside your network |
| Audit-log sink | Managed | Customer sink (S3 / Splunk / Elastic) | Customer sink (S3 / Splunk / Elastic) |
| SLA | 99.9% | 99.95%, bespoke | Your own SLA, optional support package |
| Price | From €69 / user / month | On request | On request (licence + support) |
Prices and SLA figures are indicative as of 23 April 2026; binding terms after the RFI conversation.
In practice
Already running in German university hospitals.
The EULAR 2025 study (Knitza & Aries) was conducted in a rheumatology outpatient practice: 88 consultations, documentation time reduced in 82% of cases, zero critical documentation errors, zero patient refusals. University hospitals and MVZ networks are piloting or operating Nixi AI in outpatient or ward settings. Named references shared in the RFI conversation under NDA.
Procurement
RFI, DPA, and architecture pack.
No sales call before you've seen the documents. We assemble a pack: architecture diagram, BSI C5 attestation of the infrastructure, sample data-processing agreement (§ 28 GDPR + § 203 StGB), Art. 30 GDPR data-processing register, technical, and organizational measures (TOM), and a model-card document for the open-weight models used in Dedicated or On-premise deployments.
The pack is typically provided within two business days.
Frequently asked
What hospital IT and procurement want to know.
- Is Nixi AI BSI C5-certified, or does it just run on BSI C5 infrastructure?
- The cloud infrastructure (Frankfurt) holds a BSI C5 attestation. That is the baseline § 393 SGB V has required since July 2025. Nixi's own C5 attestation is in progress. We surface the distinction deliberately because other vendors obscure it.
- How does HIS integration work for Orbis / i.s.h.med / CGM Clinical / M-KIS?
- Via FHIR R4 or HL7 v2. Nixi delivers Encounter, Observation, Condition, DocumentReference, and optionally MedicationRequest. HIS-specific adapters (Dedalus NextGen, i.s.h.med native, CGM tenant proxy, MEIERHOFER Connect) are implemented; an HIS-specific assessment is part of every enterprise onboarding.
- What are the on-premise deployment options?
- Nixi AI ships as a signed Helm chart for Kubernetes 1.27+. Three deployment shapes are supported: fully air-gapped (with an internal model registry mirror), hybrid (sensitive audio local, managed UI) and full on-premise with a dedicated identity provider (OIDC / SAML 2.0). Minimum hardware for 20 clinicians: 2× Nvidia L4 GPUs or CPU-only.
- What does a hospital pilot look like?
- Two to four weeks of pre-conversations (IT security, DPO, clinical lead), two weeks of technical onboarding (tenant setup, HIS adapter, SSO), then a four- to twelve-week pilot in a single department with a clearly defined metric (typically: documentation minutes per consultation, rate of incomplete discharge letters, MFA overtime hours). Pilot contracts are scoped to the first department; scaling follows a positive pilot review.
- EU AI Act: is Nixi AI compliant?
- Nixi AI falls under the AI Act category of 'high-risk AI' (medical application). Transparency, audit, and human-oversight requirements are implemented from the start: every AI-generated output is clinician-reviewed and approved, audit-log events are tamper-evident, and model cards are available in Dedicated and On-premise deployments. The full compliance dossier sits in the RFI pack.
Keep reading
Still documenting after your last patient leaves?
Nixi AI writes the doctor's letter while you talk. So you leave when they do.