NeMo Guardrails
Add safety, security, and control to your LLM applications NeMo Guardrails is an open-source toolkit for easily adding programmable guardrails to LLM-based conversational systems. Built by NVIDIA, it provides a flexible framework for defining rules and controls that govern how your LLM interacts with users.Get Started
Installation
Install NeMo Guardrails and set up your environment
Quickstart
Build your first guardrailed application in minutes
Core Concepts
Learn about the five types of guardrails
Examples
Explore real-world implementations
Why NeMo Guardrails?
Safety & Trust
Prevent harmful content and keep conversations on-topic
Controllable Dialog
Guide conversations along predefined paths
Secure Tool Integration
Safely connect LLMs to external services
Multi-Layer Protection
Apply guardrails at input, dialog, retrieval, execution, and output stages
Key Features
- Five Types of Guardrails: Input, dialog, retrieval, execution, and output rails for comprehensive control
- Colang DSL: Purpose-built language for defining conversational flows and guardrails
- Built-in Library: Pre-built guardrails for jailbreak detection, content safety, fact-checking, and more
- LLM Integration: Works with OpenAI, NVIDIA NIM, HuggingFace, and other providers
- Async-First: Built on Python async for high-performance applications
- LangChain Compatible: Seamlessly integrate with LangChain chains and agents