Click to Enlarge
How Generative AI Works
Generative AI models learn by recognizing patterns and structures within massive datasets of text, code, images, audio, video, and other data. These models use neural networks, often transformer networks, to process the information. Developers can then leverage the models to generate new content, enhance existing content, or create entirely new AI-powered applications. AI is only as useful as the information it can access—models trained on general data fall short when they can’t draw on domain-specific knowledge. NVIDIA Agent Knowledge Foundation connects AI agents to an enterprise’s data, systems, and context, so they can reason over the information that matters to the business and deliver accurate, relevant results.
Learn More About NVIDIA Agent Knowledge FoundationExplore Generative AI Tools and Technologies
NVIDIA Nemotron
NVIDIA Nemotron™ is a family of open and efficient multimodal models with open datasets and recipes for building agentic AI.
NVIDIA Cosmos
NVIDIA Cosmos™ is a platform of state-of-the-art generative world foundation models and data processing pipelines that accelerate the development of highly performant physical AI systems, such as robots and self-driving cars.
NVIDIA Kumo
NVIDIA Kumo is a family of open models and libraries for predictive AI over structured data. It gives agents the ability to forecast outcomes, score risk, and recommend actions from relational, tabular, and time-series data without building a custom machine learning pipeline for every question.
NVIDIA NIM
NVIDIA NIM™ is a set of easy-to-use microservices designed to accelerate the deployment of generative AI models across any cloud or data center.
NVIDIA Dynamo
NVIDIA Dynamo is an open source, low-latency inference framework for serving generative AI models in distributed environments. It scales inference workloads across large GPU fleets with optimized resource scheduling, memory management, and data transfer, and it supports all major AI inference backends.
NVIDIA NemoClaw
NVIDIA NemoClaw™ is a collection of open blueprints for building domain-specialized enterprise agents on a stack that organizations can own, customize, and run anywhere. NemoClaw combines open reasoning models, tuned agent harnesses, and NVIDIA OpenShell™ for safer, governed execution.
Manage AI Agent Lifecycle With NVIDIA NeMo
NVIDIA NeMo Anonymizer
NVIDIA NeMo™ Anonymizer detects and protects PII through context-aware replacement and rewriting. It offers user-guided entity detection, followed by flexible masking strategies, including substitution, redaction, annotation, hashing, and full-text rewriting, to preserve context while ensuring privacy.
NVIDIA NeMo AutoModel
NVIDIA NeMo AutoModel provides day-0 accelerated fine-tuning support for Hugging Face models without checkpoint conversion.
NVIDIA NeMo Compass
NVIDIA NeMo Compass is an open source library that turns stored agent traces and evaluation evidence into a prioritized improvement backlog.
NVIDIA NeMo Curator
NVIDIA NeMo Curator provides pre-built accelerated pipelines to process multimodal data at scale, improving the performance of agentic systems.
NVIDIA NeMo Data Designer
NVIDIA NeMo Data Designer generates high-quality, domain-specific synthetic data from scratch or seed examples—accelerating model development while eliminating privacy risks and data collection bottlenecks.
NVIDIA NeMo Eval Author
NVIDIA NeMo Eval Author is an open source library that designs and maintains evaluations that reflect how a specialized agent is intended and actually used.
NVIDIA NeMo Fabric
NVIDIA NeMo Fabric is an open source library that provides one consistent interface to configure, run, and collect results from agents across multiple harnesses—Claude Code, Codex, Hermes, LangChain Deep Agents, and custom harnesses. Switch harnesses without rebuilding your integration; every run returns normalized results, artifacts, and telemetry.
NVIDIA garak
NVIDIA garak is an open LLM vulnerability scanner that probes AI models and agents for hallucination, prompt injection, data leakage, jailbreaks, toxicity, and dozens of other failure modes. Point it at any model and get a structured report of where it fails.
NVIDIA NeMo Gym
NVIDIA NeMo Gym is an open library for evaluating and improving models and agents using simulated agentic environments at scale.
NVIDIA NeMo Guardrails
NVIDIA NeMo Guardrails orchestrates dialog management, ensuring accuracy, appropriateness, and security in smart applications with LLMs. It safeguards organizations overseeing generative AI systems.
NVIDIA NeMo Helix
NVIDIA NeMo Helix brings NeMo open source and third-party libraries together through one install, one CLI, and a plugin architecture. It helps developers improve and harden existing agents with repeatable evaluation, experimentation, optimization, and safety checks.
NVIDIA NeMo Megatron-Bridge
NVIDIA NeMo Megatron-Bridge serves as a powerful bridge, conversion, and verification layer between Hugging Face and the Megatron Core backend for optimized, accelerated, large-scale distributed pretraining and fine-tuning.
NVIDIA NeMo Relay
NVIDIA NeMo Relay is an open source library that traces everything an AI agent does—every tool call, model request, decision, and handoff—without replacing your existing agent framework. Use those traces to debug issues, evaluate performance, investigate behavior, and reduce inference costs.
NVIDIA NeMo RL
NVIDIA NeMo RL supports accelerated and efficient post-training and reinforcement learning of multimodal models at scale from a single GPU to thousands of GPUs.
NVIDIA NeMo Safe Synthesizer
NVIDIA NeMo Safe Synthesizer generates fully synthetic versions of sensitive tabular datasets using LLM fine-tuning with optional differential privacy, preserving the statistical properties and utility of data for downstream AI tasks while protecting sensitive information.
NVIDIA NeMo Switchyard
NVIDIA NeMo Switchyard is an open source library of routing algorithms that selects the best model for each step of an agent workflow—across any combination of open and closed models. It works out of the box with no training required and gets smarter as your agents run.
NVIDIA NeMo User Sim
NVIDIA NeMo User Sim transforms sparse or nonexistent traffic into diverse, multi-turn trajectories that power evaluation, remediation, and targeted AI training.