NVIDIA expands Agent Toolkit with Knowledge Foundation and Kumo structured open models
NVIDIA announced an expansion of its open-source Agent Toolkit at GTC Berlin on October 21, 2026, headlined by the NVIDIA Agent Knowledge Foundation and a new Kumo family of structured open models, aimed at grounding enterprise agents in company data.
The Agent Knowledge Foundation pairs NVIDIA CUDA-X libraries - cuDF, cuGraph, cuML, cuOpt and cuVS - with a new Auto-Ontology library that translates raw company data into business concepts like customers, products and orders, and maps how they relate.
NVIDIA says enterprises can adopt the foundation without moving data, rebuilding existing workflows, or changing governance controls - the three migration burdens that typically stall agent deployments on proprietary data.
The release also includes NVIDIA Nemotron Retriever and the Kumo Structured open models - Kumo Tabular, Kumo Relational, Kumo Time Series and Kumo Forecast - which help agents find relevant company information to answer questions and make predictions over relational business data.
Early adoption is already visible: NVIDIA's own supply chain organization is using the Agent Knowledge Foundation through Palantir Foundry, Ontology and AIP to deploy specialized agents, create intelligence from proprietary data, predict outcomes and optimize decisions.
Buyer read: this targets the data-grounding problem that keeps enterprise agents in pilot. The differentiator versus build-your-own RAG is the ontology layer plus hardware-accelerated data libraries; the tradeoffs are NVIDIA stack dependency and the need to validate Auto-Ontology's concept mapping against your own schema before letting agents act on it.