NVIDIA’s open-source toolkit now includes new models and libraries to help companies build secure, data-aware AI agents without moving proprietary information.
Key points
- NVIDIA added the Agent Knowledge Foundation to its open-source Agent Toolkit.
- New components include CUDA-X libraries, Nemotron Retriever, and the Kumo model family.
- Enterprises like Palantir, Databricks, and Snowflake are already using the expanded tools.
NVIDIA has expanded its open-source Agent Toolkit with new tools designed to ground enterprise AI agents in company-specific data. The update aims to help organizations customize, monitor, and govern agent behavior while maintaining strict security boundaries.
What happened
NVIDIA announced an expansion of its open-source Agent Toolkit, introducing the new Agent Knowledge Foundation. This addition allows enterprises to use their own data to inform agent answers and actions without relocating that data or altering existing governance controls. The foundation incorporates NVIDIA CUDA-X libraries, such as cuDF and cuGraph, along with a new Auto-Ontology library that translates raw data into business concepts like customers and orders.
The toolkit also adds open models including Nemotron Retriever and the new Kumo family, which covers tabular, relational, time-series, and forecasting tasks. These models help agents locate relevant information and make predictions. OpenShell provides a secure runtime boundary for controlling agent access, while NemoClaw blueprints offer reference designs for building specialized agents. A new optimization blueprint helps improve agent intelligence over time.
Companies including Palantir, Databricks, Snowflake, Amazon Web Services, and SAP are already integrating these tools. Palantir uses the foundation to create actionable intelligence from proprietary data. Databricks pairs Kumo Anomaly with its Genie Ontology for context-aware agents. Snowflake enables customers to use Kumo models in notebook environments. Amazon OpenSearch Service and ServiceNow report faster, more accurate search results using Nemotron Retriever.
Why it matters
Enterprise AI agents require strict boundaries to remain trustworthy and secure. When agents access internal company data, they must operate within defined permissions to prevent unauthorized exposure. The Agent Knowledge Foundation allows organizations to keep data in place while still enabling agents to understand and act on it.
This approach reduces the risk of data leakage and simplifies compliance with existing governance policies. By translating raw data into familiar business concepts, the tools help ensure agents interpret information correctly. Secure runtimes further limit what agents can access, giving teams control over operations. This structure supports the development of agents that are both useful and safe for production environments.
The fix
- Review current agent permissions and data access policies.
- Evaluate whether new toolkit components fit existing security frameworks.
- Monitor agent behavior logs for unexpected actions or data requests.
- Test new models in isolated environments before production deployment.
Why it matters
Teams usually start by defining clear boundaries for what agents can access and do. Check whether your current governance controls align with the new capabilities provided by expanded toolkits. Ensure that data remains within approved environments during processing. Verify that monitoring tools can track agent decisions and data usage in real time. Establish a process for reviewing agent outputs for accuracy and compliance. This guidance helps organizations integrate new AI features without compromising security or operational integrity.
What to do and how to stay safe: NVIDIA
- Audit your current AI agent permissions to ensure they follow the principle of least privilege.
- Verify that data processing occurs within approved, secure environments before adopting new tools.
- Monitor agent activity logs regularly to detect any unauthorized data access or unusual behavior.
- Consult your security team before integrating new open-source models into production workflows.
Step-by-step guide: What is password policies and how they protect systems
General security guidance from the Crypt Weekly 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 use company data to inform AI agents without moving the data or changing governance controls.
Which companies are using the expanded toolkit?
Palantir, Databricks, Snowflake, Amazon Web Services, SAP, ServiceNow, Zoom, and others are building on the Agent Knowledge Foundation.
What new models are included in the update?
The update includes Nemotron Retriever and the Kumo family, which includes Kumo Tabular, Kumo Relational, Kumo Time Series, and Kumo Forecast.
