NeMo Guardrails is a free, open-source Python library from NVIDIA for defining and enforcing safety, topic, and behavioral constraints on LLM-powered conversational systems. It supports dialog flow control, content safety via NVIDIA and Llama Guard models, jailbreak detection, PII masking, agentic tool-call validation, and hallucination mitigation. Latest release is v0.20.0 (January 2026). It runs on Python 3.10-3.13 and integrates with OpenAI, Anthropic, HuggingFace, Azure, and NVIDIA NIM. Key features: - Programmable dialog rails to control conversation flow across turns - Content safety and topic guardrails via Llama 3.1 NemoGuard 8B - Jailbreak and prompt injection detection - PII detection and masking using NVIDIA GLiNER-PII - Agentic guardrails: tool-call validation and agent workflow protection - Integrates with OpenAI, Anthropic, HuggingFace, Azure, and NVIDIA NIM
Free and open-source under Apache 2.0 license.