NVIDIA Agent Toolkit Expands, Grounding Enterprise Agents in Company Data

NVIDIA expanded its open-source Agent Toolkit with a new Agent Knowledge Foundation, built from CUDA-X libraries, a new Auto-Ontology library, Nemotron Retriever and a new family of Kumo open models for tabular, relational, time-series and forecasting work. Alongside sit OpenShell as a secure runtime and NemoClaw blueprints for building and improving agents.

The practical promise is that companies can point agents at their data without moving it, rebuilding workflows or changing governance controls. If that holds, the hardest part of enterprise agents, which is knowing what the company actually knows, gets a standard toolkit.

We think this is the right problem to attack. Most enterprise agents fail not because the model is dumb but because it doesn't understand the business: which table means customers, how orders relate to products, what a forecast should rest on. Auto-Ontology, which maps raw company data into familiar business concepts and their relationships, aims straight at that gap. Useful AI in companies mostly looks like this, unglamorous plumbing that makes answers grounded.

The partner list is long and serious. NVIDIA's own supply chain organization is using the foundation through Palantir's Foundry, Ontology and AIP. Databricks pairs Kumo Anomaly with its Genie Ontology. Snowflake is offering Kumo models in its notebooks. On retrieval, NVIDIA cites Amazon OpenSearch Service improving relevance with virtually no added latency, SAP's fine-tuned reranker running nearly 3x faster and ServiceNow cutting latency by nearly a quarter. These are vendor-reported numbers, so we treat them as directional until customers publish their own.

The strategic read is plain. NVIDIA sells the chips; now it wants the knowledge layer and the runtime that sit between agents and everything they touch. Open source lowers the adoption barrier and makes the stack feel neutral, but every CUDA-X library still runs best on NVIDIA hardware. That's not sinister, it's how platforms are built. It does mean enterprises should know which part of their agent architecture they're quietly standardizing on.

The risk side deserves equal billing. An agent that understands company data well enough to predict outcomes and suggest decisions is also an agent that can expose, misuse or misinterpret that data at scale. NVIDIA pairs the knowledge layer with OpenShell's runtime boundary for exactly that reason. I'm watching whether governance controls keep pace once these agents move from answering questions to acting on them.

The toolkit announcement was part of NVIDIA's GTC Berlin slate. Other named builders on Agent Knowledge Foundation include AWS, ClickHouse, Cloudera, Cohesity, Dassault Systèmes, Dell, Elastic, Exa, IBM, NetApp, Oracle, Perplexity, Starburst and VAST Data. Zoom is using Nemotron Retriever for document understanding and search in its AI services.