Integrate NeMo Guardrails with NemoGuard NIM Microservices

Why it matters

This tutorial belongs to the archived 25.12.0 NeMo Microservices docs, so its deployment steps are legacy. The guardrail configuration it builds has the Guardrails library’s shape, and that part still applies.

What to practise

Through the Deployment Management microservice, deploy Llama 3.3 70B Instruct and two NemoGuard NIMs — llama-3.1-nemoguard-8b-content-safety and llama-3.1-nemoguard-8b-topic-control — with NIM_ENABLE_KV_CACHE_REUSE set. Read the guardrail configuration: models entries of type: content_safety and type: topic_control on engine: nim; input flows such as content safety check input $model=content_safety beside self check input, with input and output rails both marked parallel; and the prompt task for each check. Then send the three verification requests to /v1/guardrail/chat/completions and compare what each returns.

How it fits

The model roles, $model references and prompt tasks are the ones 7.1.2 wires into a Guardrails config.yml, and running the checks in parallel is a latency choice 7.1.3 asks you to measure. Three requests prove the wiring, not detector quality; 7.1.3 needs representative traffic.