# DataInfra · RedactionEverything

A local-first document redaction workbench that finds and anonymizes sensitive information in PDFs, scans, Office files, images and text using local LLMs and vision-language models.

RedactionEverything is a local-first redaction workbench for sensitive information in real-world files such as documents, scanned PDFs, images, Word files, and plain text. It combines semantic named-entity recognition, OCR, visual feature grounding, configurable industry schemas, human review, batch processing, and export, so sensitive content can be found, reviewed, and anonymized without sending raw documents to a remote API.

Text recognition defaults to a semantic NER model (HaS Text), with regex kept only as a user-defined fallback, while visual recognition combines OCR with a visual-semantic detection service (LocateAnything-3B) and a local OpenCV detector for targets such as faces, seals, and signatures. General schemas cover people, organizations, contacts, credentials, accounts, financial values, and identifiers, with additional industry schemas for legal, finance, and healthcare use.

It is free for personal, non-commercial use under a custom license; commercial deployment, hosted service, or OEM redistribution requires a separate commercial license, and some bundled model weights carry their own non-commercial or AGPL terms.

Website: https://github.com/ttttccxxui/DataInfra-RedactionEverything