# NVIDIA NeMo Switchyard Enables Intelligent LLM Routing Through Kong AI Gateway

This session explains how NVIDIA NeMo Switchyard and Kong AI Gateway work together to route AI requests intelligently while adding governance and cost control. The central idea is that agent workflows should use a system of models rather than sending every step to a frontier model, because simpler tasks can often run on lower-cost, more efficient models. Switchyard is presented as an open-source, provider-agnostic routing library that plugs into an AI stack or gateway, reads agent signals and traces, and recommends the best model for each step using strategies such as an LLM classifier, an agent stage router, or an escalation router. Kong then acts as the enforcement layer, applying controls such as PII masking, guardrails, observability, multicloud integrations, and cost governance. A security incident response demo shows efficient models handling triage and reconciliation, while more complex reproduce and diagnose stages use frontier models, producing measurable savings versus using one model for everything.