PII Firewall

What PII Firewall is

PII Firewall intercepts personally identifiable information before it reaches an LLM, then restores original values in the response. The self-hosted, open-source project offers profiles for healthcare, finance, legal, generic, and custom domains. Detection options include regex, Presidio, GLiNER, Transformers, OpenAI Privacy Filter, hybrid configurations, and custom backends. It supports six actions for detected information: keep, pseudonymize, redact, generalize, mask, or hash. The product says it handles more than 55 languages through automatic language detection and locale-specific patterns. Integration examples include OpenAI, Anthropic, LangChain, FastAPI, streaming transport, Ollama, LiteLLM, and arbitrary HTTP LLM endpoints. Tenant-scoped token mappings and a forget operation support deleting vault mappings for a thread or case; each call also produces a trace of entity types, confidence scores, and replacements. The project is free under the Apache 2.0 license, and the maker provides a Python package. The base regex-only installation does not need ML dependencies. The maker notes that hash is intended for analytics workflows and is not used in built-in presets.

Who it suits

PII Firewall is aimed at teams handling sensitive information in healthcare, finance, or legal workflows. It may suit developers who want to self-host PII handling in LLM integrations.

Limits

The free open-source plan is a self-hosted Python package. The HASH action is for analytics workflows and is not used in built-in presets.

Getting started

1. Install the Python package provided by the maker.
2. Choose the regex-only base install if you want an installation without ML dependencies.
3. Select a built-in healthcare, finance, legal, or generic profile, or use a custom profile.
4. Choose a detection backend and an action for detected PII.
5. Use an integration example for a supported LLM, framework, streaming transport, or HTTP endpoint.

Verdict

PII Firewall provides configurable PII handling before and after LLM processing, with traces and deletion controls. Its self-hosted setup and Python package are important considerations for teams evaluating deployment.