NVIDIA has released two new tools aimed at letting companies build autonomous AI agents for engineering and cybersecurity tasks. The company launched the AI Agent Toolkit alongside the Nemotron 3 Ultra model, both designed to automate complex workflows inside large enterprises.
What the toolkit does
The AI Agent Toolkit is a suite of pre-built components. It handles common agent functions — planning, tool use, memory, and reasoning — so developers don't need to code everything from scratch. NVIDIA says the toolkit works with popular frameworks like LangChain and LlamaIndex, and it includes templates for deploying agents inside Kubernetes clusters.
The goal is straightforward: give engineers a starting point for agents that can analyze logs, write reports, or respond to security incidents. Instead of stitching together libraries and managing infrastructure manually, teams can pull from the toolkit and focus on their specific use case.
Nemotron 3 Ultra enters the picture
Alongside the toolkit, NVIDIA introduced Nemotron 3 Ultra, a language model optimized for agentic tasks. The model is built on the same architecture as the Nemotron family but tuned for function calling, multi-step reasoning, and following instructions that involve external APIs.
Nemotron 3 Ultra is designed to run on NVIDIA's own hardware, though the company also offers it through its DGX Cloud and AI Enterprise software stack. The model can handle longer context windows, which matters when an agent needs to reference multiple documents or a long chat history before taking action.
Engineering and cybersecurity workflows
NVIDIA specifically highlighted two domains where the combination of toolkit and model fits. In engineering, autonomous agents could automate parts of simulation, design review, and defect detection. Instead of waiting for a human to spot a bottleneck, an agent could monitor a pipeline and trigger fixes.
In cybersecurity, the agents could triage alerts, correlate threat intelligence feeds, and even propose containment steps. The idea is to cut down the time between a detection and a response, especially for teams that are understaffed.
The company hasn't announced a specific customer rollout yet. Developers can access the toolkit through NVIDIA's developer portal starting today. Early access to Nemotron 3 Ultra is available via the NVIDIA API catalog.
