NeMo Guardrails is NVIDIA’s open-source toolkit for adding programmable safety and control rails to LLM-based conversational applications. It has around 7,300 stars on GitHub and is in active development in 2026 as part of the NVIDIA NeMo ecosystem.
Shipping an LLM chatbot without guardrails risks off-topic answers, jailbreaks, leaked prompts and hallucinated facts. NeMo Guardrails lets developers define, in code, what an application may and may not do. Its Colang modeling language describes dialogue flows and rules declaratively: input rails screen user messages before they reach the model, dialog rails steer the conversation along approved paths, and output rails check responses before they are shown. The toolkit works with popular LLM providers and frameworks, supports techniques such as self-checking and fact-checking against retrieval sources, and can enforce topical boundaries, content policies and structured response formats. It is designed to slot into existing chatbot stacks rather than replace them, and NVIDIA maintains detailed documentation and examples. The project is free and open source.
NeMo Guardrails is free and open source. You can download it from GitHub and run it in your own infrastructure; the only costs are your existing LLM API and hosting expenses.
NeMo Guardrails is for teams deploying customer-facing chatbots, copilots and agents who need enforceable safety, compliance and quality controls around LLM behavior without building a moderation layer from scratch.