Redakto - The Incognito Tab for LLMs

Large language models (LLMs) are increasingly used in everyday applications, but ensuring privacy—especially removing personally identifiable information (PII) from text that enters an LLM—remains a major challenge. This is becoming more urgent with new EU legislation, and privacy uncertainty could slow innovation and research-to-application transfer in EU countries.

Redakto addresses this by providing state-of-the-art functions for both redaction and pseudonymization of PII. The tool is accessible to end-users through a web application and to developers and researchers via REST APIs and model context protocol (MCP) hooks. It is fully open source, requires modest compute resources, and can be deployed on local hardware.

In contrast to prior work, the authors conducted extensive empirical evaluations on legal and medical domain texts, assessing both privacy and utility of the anonymized outputs. The results show that texts anonymized with different redaction strategies achieve utility scores on par with the original texts, indicating that Redakto can be used for LLM tasks without substantial negative impact in the explored tasks.