Rizzo PII is an open-source, local-first privacy guard that detects and anonymizes personal data in documents before you share them with large language models. Built by Rizzo AI Academy and released under MIT, it has about 1,200 stars on GitHub and was updated in 2026.
Lawyers and other professionals want the productivity of frontier LLMs without sending client data to the cloud. Rizzo PII solves the first half of that problem: a compact transformer model (an mmBERT-based token classifier of roughly 0.3B parameters) finds personally identifiable information in Italian legal text across 22 categories, such as names, tax codes and case details, and replaces it with reversible placeholders. Everything runs locally and offline on ordinary CPU hardware in about half a gigabyte of RAM, with a reported micro-F1 of 0.989. After the LLM responds, placeholders can be mapped back to the original values, so the workflow stays GDPR-friendly by design. Desktop builds for Windows and macOS make it usable without any coding.
A full technical report published with the project documents the model architecture, how the training dataset was constructed for Italian legal language, and the evaluation methodology behind the reported scores, which makes the tool unusually transparent for a privacy product and easier for compliance teams to assess.
Rizzo PII is completely free and open source under the MIT license. There are no accounts, subscriptions or cloud services; the tool runs entirely on your own machine.
Rizzo PII is for lawyers, legal teams and privacy-conscious professionals who need to sanitize documents, especially Italian legal texts, before using cloud AI assistants.