Nvidia releases open weights models and toolkit to ground enterprise AI agents

What's Happening

• Nvidia introduced the Agent Knowledge Foundation at its GTC Berlin conference.
• The open-source collection adds Nemotron Retriever models, an Auto-Ontology library, and Kumo models to the company's Agent Toolkit.
• The release includes a new Nemotron Parse model trained to extract information and charts from PDF documents.
• The system allows enterprise data to remain in existing storage platforms, relying on vendors like Snowflake, Databricks, and Palantir to manage access permissions.

Why It Matters

Enterprise AI agents struggle to understand complex company data even when connected to storage systems. Nvidia's toolkit builds knowledge graphs and improves retrieval without forcing companies to move their data.