Today, we announced new tools and features for NVIDIA NeMo™ for building, training, and optimizing AI agents.
With NeMo, developers can start with open-source NVIDIA Nemotron™ reasoning models and specialize them for domain expertise—creating secure AI agents that continuously learn.
NVIDIA NeMo software now includes:
-
NeMo Data Designer for generating high-quality synthetic datasets
-
NeMo RL open-source library for scalable reinforcement learning and model alignment
-
NeMo Retriever for retrieval-augmented generation (RAG) and the NeMo Evaluator SDK for easier and faster benchmarking of AI models and agentic systems, both now open source and available on Hugging Face
-
NeMo Guardrails now supports a new multilingual content safety model and enforces AI safety and compliance at the network layer with the new NeMo Guardrails-GKE Inference Gateway integration - now available as a container on GCP Compute Engine on the Google Marketplace console to download or deploy with NVIDIA AI Enterprise license
-
NeMo Agent Toolkit an agentic framework for building and optimizing AI agent workflows, now includes new hyperparameter and prompt tuning capabilities to scale pilots to production—in less time with improved consistency
-
NeMo Curator, which now supports audio and video processing, enabling the curation of high-quality multimodal datasets for training and optimizing agents
Get started:
🖥️ Explore all the tools and techniques to get started with NVIDIA NeMo.
📖 Read the latest updates about NVIDIA Nemotron.
📅 Attending GTC Washington, D.C.? Add the Agentic AI Day sessions to your calendar.
📤 Stay up to date on agentic AI by subscribing to NVIDIA news and following NVIDIA AI on LinkedIn, X, YouTube and Discord.