Detect and redact sensitive entities from text, images, PDFs and Office documents.
Loading configuration…
Detection engine
Hybrid — Combines GLiNER neural extraction with Presidio regex/checksum recognizers. Best overall accuracy for English and French documents.
GLiNER — Zero-shot neural model. Handles any language and mixed-language documents without configuration. Supports custom entity types beyond the standard list.
Presidio — Rule-based detection using spaCy + regex patterns. Best for structured data where exact pattern matching is preferred. Requires language selection.
GLiNER confidence threshold
Lower = more entities (higher recall); higher = stricter (higher precision).
Entities to redact
Custom entity types
GLiNER-powered
GLiNER can detect any entity you describe in plain English — not just the list above.
Type a short noun phrase and press Enter or click Add.
Examples:
contract number,
employee ID,
account reference,
vehicle registration
Input type
Redaction mode
Automated detection may not catch every sensitive value — review the output before sharing.
Input type
Pseudonymized document
Key file (the .key.json file from your ZIP download)
Restored text