Security & data handling

How your content is used

When you upload a reference PDF or enter a claim, RefCheckr extracts the relevant text, compares it against your claims, and locates supporting or conflicting passages. That is the full extent of how your content is used.

Your documents and claims are not added to any training dataset, indexed for other users, or used to improve the underlying AI models.

Storage & retention

Reference Library — PDFs you save to your library are stored persistently so you can verify against them again without re-uploading. They remain there until you delete them.

Verifications — claim, verdict, and cited passages are saved to your verification history so you can return to previous results. You can delete any verification at any time.

Account deletion — when you delete your account, all of your data (library, verifications, projects, workspaces, usage, and subscription records) is removed from the live database immediately.

Backups & disaster recovery — encrypted database backups are retained on a layered schedule for disaster recovery: daily backups kept for 6 days, weekly backups kept for 1 month, and monthly backups kept for 3 months. Backups age out automatically at the end of their retention window; deleted data persists in backups only until the longest backup containing it has expired.

AI models

RefCheckr uses third-party AI models from Anthropic (Claude) and Perplexity (Sonar) to compare claims against reference text and to assess candidate references. Both providers process your input only for the duration of the request and contractually do not retain it for training.

Provider data-handling references: Anthropic · Perplexity.

Subprocessors

The following providers process customer data on our behalf. Each is bound by their standard data-processing terms and none retain content for training.

Security controls

In transit — all traffic between your browser, RefCheckr, and our subprocessors is encrypted with TLS. Connections to our database run over TLS in production.

At rest — data is encrypted at rest using our database provider's default disk-level encryption. Customer-managed encryption keys are not currently offered.

Authentication — RefCheckr uses passwordless magic-link sign-in with signed JWT session cookies. Passwords are never stored.

Access — your library, verifications, and projects are scoped to your workspace and not visible to other users. Internal staff access is limited to support and incident response.

Data location

RefCheckr's primary backend and managed PostgreSQL database are hosted on Railway in the United States. A secondary RefCheckr-operated service handling rule-based compliance checks runs on Google Cloud Run, also in the United States. AI processing requests are sent to Anthropic and Perplexity in the regions those providers operate.

If you require EU/UK data residency for an enterprise deployment, please contact us — we can scope a regional deployment as part of an enterprise engagement.

Compliance & DPA

A standard Data Processing Agreement (DPA) is available on request for enterprise customers, alongside an AI processing addendum that documents how customer content is handled by our model providers.

The RefCheckr for Word add-in is listed on Microsoft AppSource, having passed Microsoft's certification review for security and compliance.

For procurement questions, security questionnaires, or DPA requests, contact support@pharmatools.ai.