Add NeMo Guardrails to a LangGraph agent on Red Hat OpenShift AI

Add NeMo Guardrails to a LangGraph agent on Red Hat OpenShift AI without rewriting the graph. Jump to 5:28 to see the same in-cluster vLLM unguarded vs guarded: the unguarded agent answers a check-fraud prompt; NeMo Guardrails blocks it on content safety.
NeMo Guardrails is included in Red Hat OpenShift AI (upstream: NVIDIA NeMo Guardrails) and is deployed with a NeMo Guardrails custom resource managed by the TrustyAI Operator. This walkthrough covers the passthrough proxy, layered rails (regex → content safety → topic boundary → output safety), and opt-in tracing so you can see which rail fired.

Chapters
0:00 Introduction
0:17 What this agent does
1:02 Proxy architecture and rail order
2:25 Customizing rails and changing domain
4:23 Deploy locally and on OpenShift AI
4:58 Tracing overview
5:28 Unguarded vs guarded
8:04 Which rail fired
9:35 Wrap-up

Resources

Guardrailed Agent example:
https://github.com/red-hat-data-services/agentic-starter-kits/tree/main/agents/langgraph/examples/guardrailed_agent

Enable AI safety with NeMo Guardrails:
https://docs.redhat.com/en/documentation/red_hat_openshift_ai_self-managed/3.5/html/enabling_ai_safety_with_guardrails/enabling-ai-safety-with-nemo-guardrails_nemo-guardrails