Skip to content

About Guardrails

Use NeMo Guardrails to apply safety checks and content moderation to large language model (LLM) applications. Guardrails runs as an Inference Gateway (IGW) middleware plugin that evaluates user inputs and model outputs against configurable guardrail policies. It supports dedicated task models (such as content safety or topic control) and integrates directly into the inference pipeline through VirtualModels.

Guardrail configurations define which checks run, which models perform the checks, and how blocked content is handled. You wire a guardrail configuration onto a VirtualModel, and every inference request to that VirtualModel flows through the safety checks automatically — using the standard IGW OpenAI-compatible endpoint.


Concepts


Tutorials

These tutorials walk you through common guardrail tasks using NeMo Guardrails.