NVIDIA expands Agent Toolkit with knowledge and secure-runtime tools

News

NVIDIA says its open Agent Toolkit now adds knowledge, security and optimization tools to help enterprises ground AI agents in company data.

NVIDIA said on Wednesday it has expanded its open-source Agent Toolkit with new components designed to help enterprises ground AI agents in company data, control what those agents can access and monitor how they operate. The company framed the update as part of a broader enterprise stack for building agents that are both useful and governed within defined boundaries.

Evidence and context

In the Oct. 21, 2026 post, NVIDIA said the toolkit now includes the NVIDIA Agent Knowledge Foundation, NVIDIA open models such as Nemotron Retriever and the new Kumo family, NVIDIA OpenShell secure agent runtime, NVIDIA NemoClaw secure agent blueprints, and tools and skills including NeMo Relay and NeMo Compass. NVIDIA said the aim is to bring together the model, harness, runtime, data and tools needed to build agents that can perform work in production settings.

The company said Agent Knowledge Foundation is intended to let enterprises use company data without moving it, rebuilding existing workflows or changing governance controls. NVIDIA said the foundation includes CUDA-X libraries such as cuDF, cuGraph, cuML, cuOpt and cuVS, plus a new Auto-Ontology library that maps raw company data into business concepts such as customers, products and orders and shows how they relate. NVIDIA also said the foundation incorporates Nemotron Retriever and Kumo Structured open models, including Kumo Tabular, Kumo Relational, Kumo Time Series and Kumo Forecast, to help agents retrieve relevant information and make predictions.

NVIDIA described OpenShell as the secure runtime boundary for controlling what agents can access and do, while NemoClaw blueprints are presented as reference designs for specialized agents. The newest NemoClaw blueprint, NVIDIA said, is aimed at agent optimization and provides a step-by-step starting point for improving agent intelligence over time.

The company also cited enterprise adoption examples. NVIDIA said its own supply chain organization is using Agent Knowledge Foundation through Palantir Foundry, Ontology and AIP to deploy specialized agents and generate actionable intelligence from proprietary data. It said Databricks is pairing NVIDIA Kumo Anomaly with Databricks Genie Ontology, and that Snowflake customers can use the NVIDIA Kumo Engine and Kumo Relational open models in Snowflake Notebook environments. NVIDIA said Amazon OpenSearch Service improved relevance with fewer search steps and virtually no added latency using the embedding model, SAP’s fine-tuned reranker ran nearly 3x faster, and ServiceNow improved reranking quality while cutting latency by nearly a quarter.

The post follows a series of earlier toolkit expansions in 2026, including BioNeMo Agent Toolkit on June 23, Omniverse libraries on July 20 and PhysicsNeMo plus updated CUDA-X libraries on July 26. The newer release continues NVIDIA’s pattern of adding domain-specific tools to its enterprise agent platform, though the performance claims and adoption statements in the company’s post are NVIDIA-reported and are not independently verified in the material provided.