Local detection
Pattern matching and offline English and Italian language models run inside the application. No external AI provider receives the source text for detection.
SECURITY
Practical safeguards for reversible anonymization.
The application combines deterministic detection, carefully filtered language analysis, memory-only PDF uploads, protected premium access and reversible restoration.
Precision-first language analysis
Language models propose candidates, but context and confidence checks decide whether a natural-language entity is anonymized automatically.
Controlled restoration
Text restoration replaces exact placeholders from the private map; its text fingerprint is advisory. PDF restoration is position-based and intentionally uses no PDF hash or source-file fingerprint.
Memory-only PDF uploads
Multipart file streams use RAM rather than Werkzeug spooled temporary files, with a hard request-size cap. Export permanently removes selected source text before writing placeholders.
Protected paid access
Paid limits and subscriptions are validated by the server. Access-code replacement does not reset quota or create a second entitlement.
Durable migrations
Versioned backups preserve premium entitlements and usage during normal updates without including submitted text or restore maps.
Administrator protection
Repeated failed administrator sign-ins trigger a temporary lockout, and the dashboard exposes aggregate operational data rather than submitted content.
Read how the anonymization flow works, the privacy model, or the FAQ.