Security & PHI
Last updated October 3, 2026
The most common question a resident asks about any tool is: “will my program's compliance office kill it?” Here is the plain answer, and the architecture behind it — a page you can forward to your program director or GME office.
Vemri's servers do not receive, store, or transmit Protected Health Information (PHI). Vemri is not a HIPAA Business Associate. The product is architected so that patient-identifying data never reaches our backend — this isn't a policy promise layered on top, it's how the system is built.
Procedure-level, never patient-level
Vemri handles only procedure-level content: operative technique, attending preferences and pearls, danger points, and template structure. It is designed to never hold patient identifiers — no names, MRNs, dates of birth, contact information, or clinical history tied to a person — in any field, form, upload, or prompt.
Your case log: procedure-level, and yours
Vemri stores your case log itself: procedures, steps, attending preferences, pearls, pre-op plans and op-notes assembled from already-redacted dictation. It is encrypted at rest, protected by per-account access rules, and never mined for analytics. It holds no patient identifiers because nothing in the product asks for them.
Notion is optional. If you connect your own workspace, Vemri mirrors your case log into it. Reads from Notion pass through a proxy that forwards bytes straight back to your browser and never logs, caches, persists, or inspects the response.
Images are de-identified on your computer
Case images come from your own screen, a screenshot or a file. Before anything uploads, your browser re-encodes the image (dropping all file metadata), lets you crop it, and finds and removes burned-in overlay text such as names, MRNs and dates. The text it finds is discarded, never stored or sent. You then confirm the image is clean.
On upload, the server checks again: it rejects any image that still carries metadata or still looks like it has overlay text. DICOM patient and study tags are never read. Images are never sent to our AI provider or used in analytics.
De-identified by construction for AI features
A closed, opt-in set of procedure-level workflows can send text to our AI provider to help draft templates and structure your own notes. Every such request passes through an automated redaction step before it leaves, is gated behind a feature flag, and records only a metadata-only audit entry (token counts and latency). We do not log the prompt or the response, and these workflows are scoped to technique and decisions — never “tell me about the patient.”
What we never do
- No patient identifiers collected, in any surface.
- No logging of your Notion content or AI prompt/response text — not in server logs, analytics, the audit log, or error messages.
- No server-side caching or persistence of anything read from your Notion.
- No third-party advertising trackers; analytics events are count-only and carry no content.
Transport & access
All traffic is encrypted in transit (HTTPS). Authentication is handled by Supabase; billing by Stripe (we never see card details). Any credentials you choose to connect are encrypted at rest. Access to our systems is limited to what's needed to operate the service.
Sub-processors
We rely on Supabase (auth + database), Vercel (hosting), Stripe (billing), Resend (email), Anthropic (the de-identified AI workflows), and, if you connect it, your own Notion workspace. See our Privacy Policy for details.
A note on wording
We say “PHI-safe by design” — not “HIPAA-compliant” — on purpose. HIPAA compliance is a property of a covered entity's or business associate's whole program; Vemri's claim is narrower and more precise: because our servers never receive PHI, Vemri is not a Business Associate and there is no PHI for us to safeguard on your behalf. Keeping patient data out of Vemri is a shared responsibility — the app is built to prevent it, and you agree not to enter it (see our Terms).
Questions from your program
If your GME or compliance office has questions, email rserbin15@gmail.com and we'll respond directly.