guardrails-safety

What problem does it solve?

Guardrails-safety provides a structured approach to instrument safety checks, detect harmful content, and enforce policy-compliant outputs across AI agent interactions.

Core Features & Use Cases

• Input guardrails detect prompt injections and filter content before user input is processed.
• Output guardrails detect PII, hallucinations, and unsafe content, optionally redacting or blocking responses.
• Guardrail pipeline, metrics, and async guards enable scalable, observable safety across production agents.
• Use cases include defending production agents from data leakage, maintaining compliance, and reducing unsafe completions.

Quick Start

Integrate the guardrail pipeline into your agent flow to automatically enforce input and output safety checks.