Biomedical engineering and IT
Procuring AI tools for clinicians and needing to say, in writing, what leaves the network and what does not.
For hospitals
Your clinicians are already using AI. Make it safe to allow.
Tavrik catches patient details in AI traffic before they leave your network, applies the policy you set to every request, and writes the evidence into the audit systems you already run. It does not touch your clinical systems.
Who it is for
Clinical AI programme leads
Scoping a rollout from one department outward, with the evidence to extend it or to stop.
Compliance and privacy
Setting the scope of the agreement, the audit path and the records that prove policy was enforced.
What it does
Catches patient details first
37 clinical recognisers in the healthcare pack: HL7 v2, FHIR, DICOM and Australian identifiers, applied before any model sees the request.
Works in the browser
A signed browser extension covers ChatGPT, Claude, Gemini, Copilot, Perplexity, Poe, You.com, Mistral and Cursor. Nothing changes for the clinician; the submission is inspected before it leaves the page.
Works for applications
Anthropic, OpenAI, Azure OpenAI, Google Gemini, Amazon Bedrock and Ollama behind one gateway. An application repoints a base URL and a key.
Evidence in your systems
HL7 v2 messages to your integration engine, and FHIR AuditEvent resources to your audit store, plus SIEM forwarding. Categorical only: which kinds of detail, how many, what was decided. Never the text.
Where it stops
Tavrik governs AI traffic. It does not process HL7 clinical messaging between your EHR and your lab, imaging or pharmacy systems, does not mediate FHIR clinical APIs, does not replace an integration engine, and is not general-purpose data loss prevention. The boundary is stated this plainly because a tool that quietly expands into the clinical record is a different risk class.
Start with one department
Four weeks, one department, real clinicians. The onboarding plan is public; read it before you talk to us.