NVIDIA Open-Sources NeMo Switchyard for Dynamic AI Agent Routing

Sam Witteveen · youtube · 2026-08-11

NVIDIA released NeMo Switchyard, a new open-source library designed to dynamically route AI agent workloads across the most suitable models for each step of a run.

The tool supports routing decisions based on request signals, agent state, tool results, and model capabilities, cost, and latency. It offers various routing algorithms, including LLM classifiers, cascade routing, and trainable prefill routers. With built-in observability, it helps developers keep routine tasks on efficient models while switching to frontier models when necessary, bringing specialized, local, and frontier models together within a single agentic system.

Related event: NVIDIA Open-Sources Nemotron Model and Agent Router(37 posts)→

Original post →

More from coding & agent

coding & agent channel →