QuiGuard firewall stops PII leaks in agent tool calls

QuiGuard addresses a critical security blind spot where standard prompt-layer scrubbing fails to catch sensitive data leaked inside function arguments.

• –Leverages Microsoft Presidio for robust, real-time PII detection and redaction across multiple data types
• –Specifically targets tool_call JSON, ensuring agents don't inadvertently send customer data (like SSNs or emails) to external LLMs during execution
• –Provides a configurable policy.yaml allowing developers to mask, block, or warn based on granular security rules
• –Closes the loop on autonomous agent security where LLMs have direct, often unmonitored, access to sensitive internal tools
• –Minimal setup with Docker support makes it a practical addition for developers building production-grade agents with strict compliance requirements