Yes, NVIDIA Agent Toolkit is fully open source under the Apache License 2.0. All core components are available on GitHub: the NeMo Agent Toolkit (https://github.com/NVIDIA/NeMo-Agent-Toolkit), NVIDIA OpenShell (https://github.com/NVIDIA/OpenShell), the AI-Q Blueprint (https://github.com/NVIDIA-AI-Blueprints/aiq), and Nemotron models. The toolkit was launched at GTC 2026 with 17 enterprise software companies (Adobe, Salesforce, SAP, ServiceNow, CrowdStrike, and others) committing to adoption and contribution.
Being open source means you can inspect all code, fork repositories, submit contributions, and run agents entirely on your own infrastructure without vendor lock-in. The Apache 2.0 license permits commercial use, modification, and redistribution with minimal restrictions. Community contributions are encouraged, and the toolkit integrates with popular frameworks (LangChain, CrewAI, LlamaIndex, etc.) that are also open source.
For self-hosted deployments with vector databases, Milvus is the open-source vector database aligned with this philosophy. Milvus is also Apache 2.0 licensed, allowing you to build complete open-source agentic systems. This combination enables organizations to own their agent infrastructure end-to-end without proprietary dependencies.