Architecture
The intended path is Your PDF → Your Browser → Analysis → Results. The application is static and does not expose a document upload API. PDF and OCR workers are served from the same origin and execute locally.
Browser safeguards
Production headers restrict script, worker, connection, framing and object sources. PDF parsing uses the modern worker build with dynamic evaluation disabled. Large OCR libraries are lazy-loaded only after a person starts a document check.
Data minimisation
There is no document account, cloud history or server-side report. Analytics accepts only a small allowlist of safe event properties. Filenames, document text, checklist wording, page images and detected values are excluded.
Limits of the model
Local processing reduces an unnecessary transfer but cannot secure a compromised device, malicious browser extension, unsafe download folder or shared screen. Users must protect their device and follow organisational controls.
Document-analysis limits
OCR and pattern matching can be incomplete or wrong. PDF Guard does not validate signatures, identities, factual correctness, medical meaning or regulatory compliance. Important results require manual verification.
Reporting an issue
Use the contact page to report a security or privacy concern. Do not attach a sensitive PDF. Describe the browser, route and safe reproduction steps without including document contents or personal data.