Guardrails

A security layer that sits between your AI agent and the outside world, providing six tools for input validation, output filtering, and policy enforcement. It blocks prompt injection attempts through regex patterns, redacts PII like SSNs and credit cards, prevents malicious code execution, and enforces RBAC policies with rate limiting. The audit logger captures every request with metrics and supports external handlers. You'd reach for this when you need to lock down an AI agent that handles sensitive data or operates in a regulated environment. Configuration updates happen at runtime, and the policy engine lets you block specific paths, restrict dangerous tools, or enforce maintenance windows without redeploying.