NVIDIA NeMo Guardrails

NeMo Guardrails is an open-source (Apache 2.0) toolkit from NVIDIA for adding programmable guardrails to LLM-based conversational applications and agents. Introduced in 2023, it lets developers define safety and control policies that can block, alter, or validate unsafe, off-topic, malicious, or policy-violating user inputs and model responses. Its distinguishing feature is Colang, a domain-specific modeling language for describing dialog flows and rails, making NeMo the only major guardrails toolkit that also models the conversation itself rather than only filtering individual messages. The library provides several rail types: input rails, output rails, dialog rails, retrieval rails, and execution rails, supporting self-check moderation, fact-checking and hallucination detection, jailbreak and prompt-injection detection, topic control, and integration with NVIDIA's content-safety and topic-control NIM models as well as third-party moderation APIs. It integrates with LangChain and arbitrary LLM backends and ships configuration files and custom-action hooks so teams can add controls without retraining models. Distributed free as open source, NeMo Guardrails is aimed at developers and ML engineers building safe chatbots and agentic systems, and is often paired with NVIDIA's broader NIM microservice ecosystem for enterprise deployment.