Tavrik

AI governance for hospitals · Built in Western Australia

Let clinicians use AI.
Keep patient data in the hospital.

Tavrik sits inside your network. It replaces patient details before any AI tool sees them, enforces your policy on every request, and writes the evidence into the audit systems you already run. With a local model, nothing leaves the building at all.

Talk about a pilot Read how it works →

Surfaces in

  • Browser live ChatGPT, Copilot, Gemini, Claude
  • API gateway live applications and agents
  • Microsoft live Purview, Copilot Studio
  • Endpoint roadmap desktop, IDE

Tavrik · inside your network

  1. Policy
  2. Tokenise
  3. Audit
  4. Route

No inbound path from the internet. Outbound connections only.

Providers out

  • Local live Ollama, vLLM. Nothing leaves.
  • Anthropic, OpenAI, Azure, Bedrock live placeholders only, never patient data
37 clinical recognisers
in the healthcare pack: HL7 v2, FHIR, DICOM and Australian identifiers
2 clinical audit paths
HL7 v2 messages to your integration engine, and FHIR AuditEvent resources
3 deployment shapes
AU-region dedicated cloud, on-premises VM, and Kubernetes
0 lines of client code to change
the browser path needs nothing at all; the API path repoints a base URL and a key

The data path

A clinician pastes a handover note into ChatGPT. Before it leaves the hospital network, Tavrik replaces the patient details with placeholders. The AI answers normally. On the way back, the real details are restored. The provider never received them.

What the clinician typed
Mrs Eleanor Whitfield, MRN 4471902,
DOB 12/03/1948, presented with…
What the AI provider received
{NAME_1}, MRN {MRN_1},
DOB {DOB_1}, presented with…
What was recorded
request allowed · 3 items replaced
NAME, MRN, DOB · policy v3
delivered to Mirth and FHIR AuditEvent

The same policy applies whether the request came from the browser extension or from an application through the API gateway. Prompts and answers are not stored. What is recorded is that a request happened, what kinds of details were found, and whether it was allowed.

What a clinician's hospital sees

The console shows what is flowing through, what was replaced, where it was routed, and why it was allowed or blocked. Every screen below is the live product, not a mockup.

The Tavrik console Overview page
Overview. Requests governed, patient details replaced, policy denials, connector health and spend.
The Tavrik console Traffic page
Traffic. Every request, what was replaced, which model handled it, and the decision trail behind each row.
The Tavrik console Audit page
Audit. A tamper-evident chain, written in plain English, verifiable on demand.
The Tavrik console Costs page
Costs. Spend over time, this month against last, and where it goes by provider and department.

Audit into your clinical systems

Evidence goes where your existing processes already look for it, rather than into another console nobody opens.

Your integration engine

Audit events delivered as HL7 v2 messages over MLLP with TLS, to the integration engine you already run — Mirth Connect, Cloverleaf, Rhapsody and Iguana are the common ones — and your integration team routes them like any other feed.

Your FHIR estate

FHIR R4 AuditEvent resources over HTTPS. Audit records only. Tavrik does not read or write clinical resources.

Your SIEM

Splunk HEC, Microsoft Sentinel and Microsoft Purview. The same categorical events, in the tools your security team already watches.

How the audit path fits a hospital →

Where it runs

Dedicated single-tenant cloud in your region, an on-premises VM in your data centre, or your Kubernetes cluster. Inside your network either way, with the option to route to a local model so patient data never leaves at all.

Straight answers

Certification

We maintain control mappings for SOC 2, ISO 27001, HIPAA, the Australian Privacy Act and the EU AI Act. We do not yet hold a SOC 2 or ISO attestation; a SOC 2 Type I is in progress.

Customers

Tavrik is founder-led, and we are taking on our first hospital design partner. There are no public reference customers yet.

Everything here

Every claim on this page is listed with its evidence in the repository, and the build fails if a number appears without one. Ask and we will demonstrate it live.

The security posture in full →

The pilot programme

Four weeks, one department, real clinicians. We install inside your network, enrol a pilot group on the browser extension, connect the audit path to your integration engine or FHIR store, and hand you the evidence at the end. You keep the audit log either way.

Talk about a pilot