NVIDIA Agent Toolkit Expands, Grounding Enterprise Agents in Company Data
NVIDIA expands its open‑source Agent Toolkit with the Agent Knowledge Foundation, adding CUDA‑X libraries, Auto‑Ontology, Nemotron Retriever and the Kumo family of structured models to let enterprises ground agents in proprietary data without moving it. The foundation includes secure runtime (OpenShell) and blueprints (NemoClaw) that enable step‑by‑step agent optimization, already deployed at scale with partners like Palantir, Databricks, Snowflake and Zoom. Engineers can now integrate these components to build data‑aware, secure agents that respect existing governance and workflow constraints.
⚡ Key Takeaways
- Nemotron Retriever models enable search that is “nearly 3x faster” and reduce latency by “nearly a quarter” in enterprise deployments.
- The foundation bundles CUDA‑X libraries (cuDF, cuGraph, cuML, cuOpt, cuVS) plus the Auto‑Ontology library to translate raw data into business concepts.
- Secure runtime is provided by OpenShell, while NemoClaw blueprints give a concrete optimization pathway for agents.
- Integration points include Nemotron Retriever, Kumo Tabular/Relational/Time Series/Forecast models, and the Auto‑Ontology library for data grounding.
- Prerequisite: enterprises must have their data in place; the foundation works without moving data or altering existing governance controls.
By delivering faster, lower‑latency retrieval and a turnkey secure stack, NVIDIA’s toolkit lets ML engineers ship production agents that can safely leverage internal data, accelerating AI‑driven decision making while meeting compliance requirements.
✅ Practical Steps
- Incorporate the NVIDIA CUDA‑X libraries (cuDF, cuGraph, cuML, cuOpt, cuVS) and the Auto‑Ontology library into your data preprocessing pipeline to generate business‑level entities.
- Deploy Nemotron Retriever and the appropriate Kumo structured model (e.g., Kumo Relational) within the OpenShell runtime to enable secure, low‑latency retrieval for your agents.
- Apply the latest NemoClaw blueprint for agent optimization to iteratively improve agent intelligence and performance.
Want the full story? Read the original article.
Read on NVIDIA Blog ↗