CloakLLM
Patient data stays home when clinical teams use AI
CloakLLM sits between an application and its language model. It finds personal data in a prompt — patient names, dates of birth, insurance and record numbers, addresses, contact details — and replaces it with reversible tokens before the prompt ever leaves the hospital's or sponsor's own infrastructure. The model still sees the clinical content it needs; it never sees who the patient is. For documentation, triage support, coding or trial paperwork that means a team can use a commercial model without sending identifiable health data to a third party. The audit log is hash-chained, timestamped and holds none of the original data, so the interaction records the EU AI Act asks for under Articles 12 and 19 exist without storing the personal data the GDPR tells you to minimise, and an auditor can verify them on their own machine with a standalone tool, without our code. MIT-licensed and self-hosted: no prompt ever reaches us. It is a tool for producing evidence, not a medical device and not a guarantee of compliance.
Go to website