Detection categories

Northdocs has 34 built-in categories, plus manual boxes, including names, addresses, emails, phone numbers, national IDs (such as CPR), company numbers (CVR), bank/IBAN details, health information, political opinions, trade union membership, criminal records and offences, child welfare and placement cases, titles, confidential markings, attorney-client privilege, and attorney work product.

How finding works

• Structured data (IDs, emails, card numbers, IBAN, CPR-style numbers) is matched with deterministic patterns. IBAN (mod-97) and card numbers (Luhn) are checksum-validated; CPR numbers are checked for a plausible date of birth, not a checksum.
• Unstructured data (names, organizations, addresses, health narrative, privilege language) uses on-device NER plus optional OVHcloud AI Endpoints after first paint.
• Political opinions and trade union membership (GDPR Art. 9), criminal records (Art. 10), and child placement cases (databeskyttelsesloven § 8, Barnets Lov) are found by AI on pages that use that language, and only when the text ties them to a person. Explicit statements about one person, such as 'medlem af 3F', 'hans straffeattest', 'dømt for vold', and 'Barnet er anbragt uden for hjemmet', are found on first paint without AI. Policy and law text about people in general is left to the AI pass.

Tune what gets flagged

You can enable or disable categories under Settings → Categories, and add keyword lists under Settings → Keywords to always redact (include) or never flag (exclude) specific terms. Magic Redact (available on Pro and above) describes a concept in plain language when a pattern is not enough.