Nvidia's NeMo Agent Toolkit: Making Agents Reliable

Join this new short course on Nvidia’s NeMo Agent Toolkit, taught by Brian McBrayer, Solutions Architect in Generative AI at Nvidia.

Many teams struggle to turn agent demos into reliable systems that are ready for production. Nvidia’s open-source NeMo Agent Toolkit (NAT) provides the building blocks you need to harden your agents for production, whether built in raw Python, LangGraph, CrewAI, or any other framework.

NAT makes it easy to add observability, run systematic evaluations, and deploy with production features like authentication and rate limiting. In this course, you’ll build a climate data analysis agent using configuration-driven workflows, add OpenTelemetry tracing to debug agent reasoning, measure performance improvements, and deploy with a professional interface. You’ll also expand to multi-agent workflows where specialized agents built with different frameworks collaborate on complex tasks.

In detail, you’ll:

• Build your first NAT workflow using configuration-driven development by creating a climate science chatbot, running it locally, and serving it as a REST API with minimal code.
• Add intelligent features by registering Python functions as tools for analyzing NOAA climate data, defining input schemas with Pydantic, and transforming your chatbot into a ReAct agent.
• Enable observability with Phoenix tracing to visualize agent reasoning, tool selection decisions, and identify performance bottlenecks for confident debugging.
• Integrate multiple agents from different frameworks by combining NAT agents with LangGraph subagents using NAT’s framework-agnostic orchestration.
• Evaluate and improve performance with NAT’s evaluation framework by creating gold-standard datasets, uncovering bugs, and making data-driven improvements through configuration experiments.
• Deploy to production with NAT UI by configuring authentication and rate limiting, enabling caching, and launching your workflow with REST APIs and a professional web interface.

NAT works directly with frameworks you already use to make agents that are observable, measurable, and deployable. Start building agents that deliver reliable results.

AI builders who want to make their agents production-ready. Basic familiarity with Python and LLM application development is recommended to make the most of this course.