Open Models and Libraries
Choose the components required for your workflow. Each works with existing data platforms, ontologies, vector databases, and agent frameworks; they do not require adopting a new end-to-end platform.
Reliable Agents: More Accurate, Grounded Answers
NVIDIA Nemotron Retriever
Open, leaderboard-winning models that improve agent context and lower agentic retrieval by 40%. Nemotron Parse processes multimodal documents up to 3.5x faster; Nemotron Embed produces 24% fewer incorrect answers; Nemotron Rerank improves retrieval accuracy by approximately 5%.
NVIDIA Kumo
Open, pretrained foundation models for making predictions with leading accuracy directly from relational, tabular, and time-series data. Kumo gives agents predictive context from structured data, enabling them to forecast outcomes, rank opportunities, recommend actions, and assess risk.
NVIDIA Auto Ontology
Generates a starting ontology from enterprise data and documentation that business experts can test, refine, and certify. It complements existing ontology platforms and reduced ontology creation time by 90% in NVIDIA testing.
More Performant Agents: Faster Access to Knowledge and Insights
NVIDIA cuDF
Accelerates the loading, joining, filtering, and transformation of structured data used by agents through familiar tools such as pandas, Polars, Spark, and DuckDB—by up to 5x on supported workloads.
NVIDIA cuML
Accelerates familiar classification, regression, clustering, and anomaly-detection algorithms for custom machine-learning pipelines—by up to 50x on supported workloads.
NVIDIA cuOpt
Turns predictions, objectives and business constraints into optimized routes, schedules, and resource allocations—up to 10× faster on supported optimization workloads.
NVIDIA cuVS
Accelerates the vector indexing and similarity search used to retrieve context for agents, with integrations for FAISS, Milvus, and OpenSearch—by up to 18x on supported workloads.
NVIDIA cuGraph
Accelerates analysis of relationships across customers, products, transactions, and other connected enterprise data, giving agents additional context for analysis and reasoning.
Click to Enlarge
From Company Data to Company Intelligence:
Prepare: Use Nemotron Parse to preserve document content and structure, and cuDF to prepare structured data.
Retrievel: Use Nemotron Embed and cuVS to find relevant information, then Nemotron Rerank to prioritize the best evidence..
Understanding: Use Auto-Ontology to generate a starting ontology that business experts can test, refine, and certify.
Predict: Use cuGraph to analyze connected data, cuML for custom machine-learning pipelines, and Kumo for pretrained prediction over structured data.
Optimize: Use cuOpt to turn predictions, objectives, and constraints into optimized routes, schedules, and allocations.
NVIDIA Agent Knowledge Foundation Use Case Examples
These technologies provide individual value and can be adopted independently or used together for higher-performing enterprise agents with reliable answers, predictions, and optimized decisions at scale.
Supply Chain
Tools: cuOpt, cuVS, Nemotron
Developer Example:
Algorithmic Trading
Tools: cuOpt, cuML, Nemotron
Developer Example: Portfolio Optimization
Developer Example: AI Model Distillation for Financial Data
Machine Learning Pipelines
Tools: cuML, Nemotron
Notebook: Data Science ML Agent