What problem does it solve? LLM applications in production face jailbreak attempts, toxic inputs, hallucinated outputs, and PII leakage. This Skill provides programmable runtime guardrails that intercept and validate inputs and outputs before they reach users or the model. ## Core Features & Use Cases - Jailbreak & Injection Detection: Block prompt injection and jailbreak patterns before they reach the LLM using Colang flows. - Self-Check Input/Output Validation: Register custom actions for toxicity scoring, hallucination detection, and fact-checking with retrieval. - PII Filtering & Integrations: Mask sensitive data with Presidio and integrate LlamaGuard for moderation on input and output rails. - Use Case: A customer support chatbot needs to refuse illegal requests, mask user SSNs, and verify factual claims in responses. Define Colang flows for each check and wrap the LLM with LLMRails to enforce them at runtime. ## Quick Start Install nemoguardrails with pip and ask the assistant to create a Colang rail configuration that blocks jailbreak attempts from your chatbot.
