NVIDIA Agent Toolkit Expands for Enterprise Agents

NVIDIA has expanded the open-source Agent Toolkit with new tools that enable companies to customize and control their agent systems. This toolkit is crucial for enterprises looking to develop agents that leverage business-specific data for optimization and decision-making. The toolkit includes the NVIDIA Agent Knowledge Foundation, CUDA-X libraries, open models such as Nemotron Retriever and the new Kumo family, and various blueprints like NVIDIA NemoClaw.

The NVIDIA Agent Knowledge Foundation allows companies to transform raw data into actionable business insights that agents can use. This not only helps agents find relevant information but also enables predictive analytics and supports decision-making. Integrating this foundation does not require restructuring existing workflows and maintains governance controls. Among these are the new Auto-Ontology library that translates business data into recognizable concepts such as customers and products.

The open-source platform also provides a secure runtime environment via OpenShell, allowing enterprises to exert control over what agents can achieve. The accompanying NemoClaw blueprints are guidelines for building secure specialized agents, with the latest blueprint aimed at optimizing agent intelligence. This helps agents improve their performance in their tasks as they gain more experience.

Several prominent companies such as Palantir, Databricks, and Snowflake are already utilizing agents from NVIDIA's expertise to generate valuable insights and make better decisions. The use of Nemotron Retriever models has improved the speed and accuracy of search processes at companies like Amazon and Zoom. Many other technology companies, including AWS and IBM, are also building on the capabilities of the Agent Knowledge Foundation.

Read the full article from Nvidia.