AI Guardian Documentation

AI Guardian is an OpenAI-compatible security proxy that sits between your application and any LLM endpoint. It analyzes every request and response for threats, routes them based on a configurable risk policy, and provides a human-in-the-loop review queue for edge cases.

Architecture

At its core, AI Guardian is a FastAPI application that implements the same POST /v1/chat/completions interface as OpenAI. Your application sends requests to AI Guardian instead of directly to OpenAI. AI Guardian then:

1. Filters the input — scans messages for 165+ detection patterns across 25+ threat categories and computes a risk score (0–100).
2. Routes based on policy — auto-allow (safe), queue for human review (ambiguous), or auto-block (dangerous).
3. Filters the output — scans LLM responses for PII leaks, API keys, and other sensitive data before returning them.
4. Logs everything — every request is written to the audit log with full metadata.

Quick Links

Support

For bugs and feature requests, open an issue on GitHub. For enterprise support, email ueda.bioinfo.base01@gmail.com.