Presidio

Presidio is made of four modules. The analyzer finds personal data in text, such as names, locations, phone numbers, credit card numbers and social security numbers, using recognizers built on named entity recognition, regular expressions, rule-based logic and checksums, with surrounding context taken into account. The anonymizer then replaces, masks, redacts or encrypts what was found. The image redactor does the same for text in images, including DICOM medical images, through OCR, and the structured module handles tables and semi-structured data.

It is used in LLM applications to remove personal data from a prompt before it is sent to a model, and the docs include a sample that masks LLM calls through a LiteLLM proxy. Recognizers can be added for new entity types, languages and NLP models, and external detection models can be connected. It runs as Python packages, in PySpark, or as HTTP services from Docker images on Kubernetes.