NVIDIA NeMo Guardrails

NVIDIA NeMo Guardrails

• Introduction
• Documentation
• Installation Guide

Getting Started

• Hello World
  • Prerequisites
  • Step 1: create a new guardrails configuration
  • Step 2: load the guardrails configuration
  • Step 3: use the guardrails configuration
  • Step 4: add your first guardrail
  • CLI Chat
  • Server and Chat UI
  • Next
• Core Colang Concepts
  • Prerequisites
  • What is Colang?
  • Core Concepts
  • Guardrails
  • How does it work?
  • The follow-up question
  • Wrapping up
  • Next
• Demo Use Case
  • Next
• Input Rails
  • Prerequisites
  • Config Folder
  • General Instructions
  • Sample Conversation
  • Testing without Input Rails
  • Jailbreak Attempts
  • Using the Input Rails
  • Testing the Bot
  • More on Input Rails
  • Next
• Output Rails
  • Prerequisites
  • Output Moderation
  • Using the Output Checking Rail
  • Custom Output Rail
  • Test
  • Next
• Topical Rails
  • Prerequisites
  • Topical Rails
  • Wrapping Up
  • Next
• Retrieval-Augmented Generation
  • Prerequisites
  • Usage
  • Wrapping Up
  • Next

Colang 2.0

• Overview
  • Motivation
  • Colang 1.0
  • Colang 2.0
  • Interaction Model
• What’s Changed
  • Terminology
  • Syntax Changes
  • Defining User and Bot Intents
  • Flow naming conventions
  • The Generation Operator
  • Active Flows
  • Entry Point
  • Import Mechanism
  • Standard Library
  • Asynchronous Actions
  • Naming Conventions
  • Multi-modal
  • Variables
  • String formatting
  • LLM invocation
  • Python API
• Getting Started
  • Prerequisites
  • Terminology
  • Guides
• Language Reference

User Guides

• Configuration Guide
  • Custom Initialization
  • General Options
  • Guardrails Definitions
  • Exceptions
  • Knowledge base Documents
• Guardrails Library
  • LLM Self-Checking
  • Community Models and Libraries
  • Third-Party APIs
  • Other
• Guardrails Process
  • Overview
  • Categories of Rails
  • The Guardrails Process
  • The Dialog Rails Flow
• Colang Guide
  • Why a New Language
  • Concepts
  • Syntax
  • Conclusion
• LLM Support
  • Evaluation experiments
  • LLM Support and Guidance
• Python API
  • Basic usage
  • RailsConfig
  • Message Generation
  • Actions
  • Action Parameters
• CLI
  • Guardrails CLI
  • Options
• Server Guide
  • Guardrails Server
  • Actions Server
• LangChain
  • LangChain Integration
  • RunnableRails
  • Chain-With-Guardrails
  • Runnable-As-Action
• Detailed Logging
  • Output Variables
• Jailbreak Detection Heuristics
  • Using Jailbreak Detection Heuristics
• NVIDIA AI Endpoints
  • Using LLMs hosted on NVIDIA API Catalog
• Vertex AI
  • Using LLMs hosted on Vertex AI
• Multi Config API
  • Multi-config API
• Migrating from Colang 1 to Colang 2
  • How it Works
  • Usage
  • Assumptions and Limitations

Security

• Security Guidelines
  • The Golden Rule
  • Assumed Interaction Model
  • Specific Guidelines
  • Adversarial testing
  • Conclusion
• Red Teaming
  • Overview
  • Challenges
  • Rating
  • Recording the results

Evaluation

• LLM Vulnerability Scanning
  • Garak
  • Scan Results
  • LLM Vulnerability Categories

Advanced User Guides

• Generation Options
  • Output Variables
  • Additional LLM Parameters
  • Additional LLM Output
  • Detailed Logging Information
  • Disabling Rails
  • Limitations
• Prompt Customization
  • Task-oriented Prompting
  • Prompt Configuration
  • Predefined Prompts
  • Custom Tasks and Prompts
• Embedding Search Providers
  • Batch Implementation
  • Custom Embedding Search Providers
• NeMo Guardrails with Docker
  • Introduction
  • Prerequisites
  • Build the Docker Images
  • Running using Docker
  • AlignScore Fact-checking
• Streaming
  • Configuration
  • Usage
• AlignScore Deployment
• Extract User-provided Values
  • Overview
  • Single Values or Lists
  • Multiple Values
  • Contextual Queries
• Bot Message Instructions
• Event-based API
  • Event Types
  • Custom Events
  • Typical Usage
• Jailbreak Detection Heuristics Deployment
  • Running on GPU
• Self-hosting Llama Guard using vLLM
• Nested AsyncIO Loop
• Vertex AI Setup

Other

• Architecture
• Glossary
  • Recommended naming conventions
• Frequently Asked Questions (FAQ)
  • Table of Contents