To get started, clone the NeMo Agent Toolkit repository from GitHub and install dependencies using Python 3.11 or 3.12 with uv sync. The toolkit requires Git, Git LFS, and the uv package manager. After cloning, run git lfs install/fetch/pull to download datasets, then activate the virtual environment with uv sync --all-groups --all-extras.
Verify installation by running the nat command to check the version and access help. The official NVIDIA NeMo Agent Toolkit documentation includes Quick Start examples showing how to profile existing LangChain or LlamaIndex agents, configure tool-calling agents, and run evaluation workflows. Optional dependencies can be installed per framework—for LangChain support, use nvidia-nat[langchain].
For vector database integration, configure Milvus as your retrieval backend by specifying host and port parameters in your RAG configuration. Milvus documentation provides step-by-step RAG setup guides compatible with the toolkit’s LangChain integration.