NVIDIA Adds Knowledge Foundation to Agent Toolkit for Enterprise Data Use

NVIDIA has expanded its open source Agent Toolkit with a new Knowledge Foundation that allows enterprises to ground AI agents in their own proprietary data. According to NVIDIA blog posts, this addition helps developers customize, control, and monitor agents while keeping them within defined operational boundaries. The expansion aims to make agents more useful by connecting them directly to company intelligence rather than relying solely on general training data.

The risk

The primary challenge addressed by this update is ensuring that AI agents remain trustworthy and accurate when interacting with sensitive business information. NVIDIA states that agents are only as useful as what they know and only as trustworthy as the boundaries they run within. Without proper grounding in specific company data, agents may produce irrelevant or incorrect outputs. The new toolkit components are designed to help builders create agents that accomplish real work while maintaining governance controls throughout development and deployment.

Who is exposed

The NVIDIA Agent Toolkit is intended for software developers and enterprises building specialized AI agents. Leading companies are already using the expanded toolkit to customize and control their systems. According to NVIDIA, Palantir is using the foundation through its Foundry platform to deploy agents that create actionable intelligence from proprietary supply chain data. Databricks pairs NVIDIA Kumo Anomaly with its Genie Ontology to power context-aware agents, while Snowflake enables customers to use Kumo models in notebook environments for predictive capabilities.

Cost and compliance

NVIDIA emphasizes that companies can use the Agent Knowledge Foundation without moving data, rebuilding existing workflows, or changing governance controls. This approach supports compliance by keeping data within existing enterprise environments. The foundation includes OpenShell, which provides a secure runtime boundary to give enterprises control over what agents can access and do. NVIDIA NemoClaw blueprints offer reference designs for building secure, specialized agents, including a new blueprint for optimizing agent intelligence over time.

What should you do?

Developers should review the new NVIDIA Agent Knowledge Foundation documentation to understand how it integrates with existing data workflows. Teams should evaluate whether their current agent architectures benefit from the included CUDA-X libraries and open models. Organizations should assess their governance controls to ensure that the new runtime boundaries align with their security requirements before deploying specialized agents in production environments.

The background in plain language

AI agents are software programs that can perform tasks autonomously, but they need access to relevant information to be effective. "Grounding" an agent means connecting it to specific, up-to-date data sources rather than relying only on its initial training. This ensures responses are accurate and contextually appropriate for the organization. Secure runtime boundaries act like guardrails, defining what data an agent can access and what actions it is permitted to take. These controls help prevent unauthorized access or erratic behavior, ensuring that automated decisions remain within acceptable operational limits defined by the enterprise.

Background: AI agents

AI agents are software systems that perceive their environment, make decisions, and execute actions to achieve specific goals. Unlike standard models, they maintain internal state and interact with external tools, creating unique security risks around tool execution, memory retention, and autonomous decision loops that require specialized controls.

Read the full guide: AI Agents Explained: Answers to Security and Architecture Questions

What to do and how to stay safe: NVIDIA

• Review the NVIDIA Agent Toolkit documentation to understand how the Knowledge Foundation integrates with your existing data pipelines.
• Evaluate whether your current agent implementations require the specific CUDA-X libraries or open models included in this update.
• Assess your organization's governance controls to ensure that the new runtime boundaries align with your security and compliance requirements.
• Monitor partner implementations, such as those by Palantir or Databricks, for best practices in deploying grounded agents in production.

Step-by-step guide: Browser security settings are the architectural walls of your digital life

General security guidance from the Vector Tribune newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is the NVIDIA Agent Knowledge Foundation?

It is a new component of the NVIDIA Agent Toolkit that helps enterprises turn proprietary data into intelligence that AI agents can use for predictions and decision-making.

Which open models are included in this expansion?

The expansion includes NVIDIA Nemotron Retriever and the new Kumo family, which consists of Kumo Tabular, Kumo Relational, Kumo Time Series, and Kumo Forecast models.

How does the toolkit help with data security?

OpenShell provides a secure runtime boundary that gives enterprises control over what agents can access and do, while NemoClaw blueprints offer reference designs for building secure agents.