nemo-guardrails

Detect and block unsafe LLM interactions using Colang 2.0 rail definitions.

Updated Feb 15, 2026

Search Tags:#colang #jailbreak detection #pii filtering #fact checking #llm safety #toxicity detection

Ready to run

npx skills add https://github.com/ricable/mcai --skill nemo-guardrails-ricable

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: nemo-guardrails
Source: https://github.com/ricable/mcai/tree/main/.agents/skills/nemo-guardrails
Command: npx skills add https://github.com/ricable/mcai --skill nemo-guardrails-ricable

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

NeMo Guardrails helps teams add runtime safety controls to LLM applications so unsafe, toxic, private, or unverified outputs are filtered before they reach users.

Core Features & Use Cases

Jailbreak detection: Block prompt injection and override attempts before they reach the model.
Input and output validation: Check user messages and model responses for toxicity, policy violations, or hallucinations.
PII and fact-checking workflows: Mask sensitive data and verify claims using custom actions and external integrations.
Use case: Protect a production chatbot by refusing malicious requests, redacting personal data, and stopping unsupported answers automatically.

Quick Start

Ask the skill to create a basic NeMo Guardrails flow that blocks jailbreak prompts and validates user input for your LLM app.

Frequently Asked Questions about nemo-guardrails

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I add runtime safety to my LLM app to block jailbreak prompts?▼

To add runtime safety and block jailbreak prompts, you can use NeMo Guardrails to detect and filter unsafe LLM interactions before they reach the model. It requires Colang 2.0 rail definitions and Python action hooks to intercept malicious requests.

What is the best way to filter PII and detect toxicity in LLM outputs?▼

Filtering PII and detecting toxicity in LLM outputs involves validating model responses using specialized rails. NeMo Guardrails supports input and output validation workflows, integrating with tools like Presidio to mask sensitive data and block toxic content automatically.

How does fact checking and hallucination detection work for production chatbots?▼

Fact checking and hallucination detection for production chatbots work by validating model responses against established rails before output. NeMo Guardrails uses custom verification logic and external integrations to stop unsupported answers from reaching users.

Can I use Presidio and LlamaGuard for input validation in NeMo Guardrails?▼

Yes, you can use Presidio and LlamaGuard for input validation in NeMo Guardrails. The framework supports optional integrations with these tools alongside custom verification logic to handle PII masking and content safety.

Do I need Colang 2.0 to implement LLM safety and output validation?▼

Yes, you need Colang 2.0 rail definitions to implement LLM safety and output validation with NeMo Guardrails. These rail definitions configure the runtime safety controls and structure the Python action hooks required for filtering.