The NVIDIA NeMo Agent Toolkit is an open-source library for connecting and optimizing teams of AI agents, published on GitHub with about 2,700 stars. It works across agent frameworks, letting you compose agents, tools and workflows built with different stacks into one system.
Agent applications are usually locked into whichever framework they were built with. The NeMo Agent Toolkit provides a framework-agnostic layer: agents from LangChain, Llama Index, CrewAI or custom code can be registered as functions, wired into workflows, and exposed through a standard server, while built-in profiling and evaluation tools measure latency, token usage and quality so you can optimize the whole system rather than guessing. Observability integrations and an experiment-tracking workflow support production hardening. The toolkit is Apache-2.0 licensed and actively developed by NVIDIA in 2026. The repository includes example workflows, a UI for inspecting agent runs and migration guides, and it is developed in the open with releases published to PyPI as the nvidia-nat package.
The NeMo Agent Toolkit is completely free and open source under the Apache-2.0 license. There are no paid tiers; you run it on your own infrastructure and pay only for the models and services your agents call.
The NeMo Agent Toolkit is for AI platform engineers and agent builders who combine multiple frameworks and need to profile, evaluate and optimize the combined system.