PII Firewall

PII Firewall is an open-source privacy firewall for LLM applications. It intercepts personal data before it reaches OpenAI, Anthropic, or other LLMs, then rehydrates it in the response. The tool is described as a transparent privacy layer between an application and any LLM, with zero changes to existing prompt logic.

Its workflow is organized around detection, anonymization, and rehydration. The page shows that one or more backends, including regex, Presidio, GLiNER, and Transformers, detect entities, and domain rules decide what to keep. Entities can be handled by disposition as keep, pseudonymize, redact, generalize, mask, or hash. The examples on the page show healthcare usage that keeps diagnoses and other clinical context while stripping identifiers such as names, national IDs, email addresses, ages, and account data. The tool also advertises 55+ languages and domain-aware profiles, with built-in presets for healthcare, finance, legal, and generic or custom use.

The documentation presents it as usable in a few lines of code. The quick start shows a Python import, create_firewall(), and a firewall.process() call, and the text states that the sanitized prompt can be forwarded to OpenAI, Anthropic, Mistral, or local models. The backend section describes several detection options: regex for structured IDs and fast pipelines, Presidio for general-purpose NER, GLiNER for zero-shot entity types, Transformers for domain-specific accuracy, and an OpenAI Privacy Filter. Each backend is shown with an install command using pip and a firewall call.

PII Firewall is labeled Open Source and Apache 2.0. The page does not identify a company maker, but it does link to GitHub and documentation. It is a Python-based privacy layer for LLM apps rather than a general-purpose AI model or chat product.