Nvidia said in its 21 October announcement that it had expanded its open-source Agent Toolkit with tools for connecting enterprise agents to company data and controlling what they can do. The release was part of its GTC Berlin 2026 announcements.
The new Agent Knowledge Foundation includes an Auto-Ontology library that, according to Nvidia, turns raw data into business concepts such as customers, products and orders, then maps the relationships between them. Think of it as labelling records and drawing the links among them: an agent could use those links to find information relevant to a question. Nvidia says companies can use the foundation without moving their data or rebuilding existing workflows.
Looking up an order could become a question about its relationship to a customer or product, where the company data contains those records and connections.
The toolkit also adds Kumo models for tabular, relational and time-series data, and forecasting. Nvidia says its supply-chain organisation is using the foundation through Palantir’s Foundry, Ontology and AIP products to deploy specialised agents. In search work, Nvidia says ServiceNow improved reranking quality while cutting latency by nearly a quarter using Nemotron Retriever.
