QuiGuard

QuiGuard is a self‑hosted reverse proxy that intercepts traffic between AI agents and large‑language‑model providers. It automatically detects and redacts personally identifiable information using Microsoft Presidio NER, applies deterministic hash placeholders for lossless restoration, and enforces guardrails that block data exfiltration, prompt‑injection attempts, and unsafe tool commands. The proxy operates without any code changes to the existing AI stack and supports a wide range of LLM models and providers.

The system includes a no‑code policy editor that lets administrators enable or disable specific PII entity types, adjust confidence thresholds, add custom regular expressions, and configure domain‑blocking or URL‑sanitization rules. A real‑time compliance audit ledger records each request, providing searchable logs and forensic analysis that can be exported for GDPR, HIPAA, or SOC 2 reviews. Multi‑tenant API keys with plan‑based access control isolate users, and the dashboard presents summary cards and timeline charts for monitoring.

QuiGuard is released under an MIT license, is positioned as experimental, and is designed for security teams that need an easy‑to‑deploy, provider‑agnostic solution for protecting AI agent data and ensuring regulatory compliance.