NVIDIA has introduced a new open-source software toolkit aimed at making enterprise AI agents safer and easier to deploy. Announced at GTC 2026 in San Jose, this toolkit addresses a key challenge: how to give AI agents enough freedom to act while keeping control over data and liability. It’s designed for developers and companies looking to build autonomous AI systems that can operate inside complex enterprise environments without risking security or compliance issues.

OpenShell and the Safety Framework

The core of the NVIDIA Agent Toolkit is called OpenShell, an open-source runtime that enforces security and privacy rules for autonomous AI agents. In NVIDIA’s terms, these agents are called “claws,” and OpenShell acts as their guardrail, making sure they follow set policies. This setup helps companies trust their AI agents to perform actions without overstepping boundaries.

Jensen Huang highlighted the importance of this technology at GTC, saying that OpenShell and related tools are pushing AI beyond simple tasks like reasoning and generation into real-world action. He explained that teams will be able to deploy specialized agents to enhance productivity while maintaining control over how they interact with enterprise systems.

NVIDIA is partnering with major security companies like Cisco, CrowdStrike, Google, Microsoft Security, and TrendAI to integrate OpenShell into their security tools. This collaboration aims to create a robust safety net for autonomous agents, ensuring they operate securely within various enterprise infrastructures.

Cost-Effective Search and Broader Adoption

The toolkit also includes NVIDIA AI-Q, a search blueprint built using LangChain. It combines high-level models for managing tasks with NVIDIA’s Nemotron models for intensive research. This hybrid approach helps reduce query costs by over half, while still delivering high accuracy. This is especially important for enterprises worried about AI expenses ballooning at scale, after initial pilot tests.

By making AI more affordable and predictable in cost, NVIDIA hopes to accelerate adoption across industries. Companies are often cautious about AI spending, especially when costs spike unexpectedly once they move beyond small-scale pilots. The new toolkit aims to address those concerns by offering a more predictable, cost-efficient solution.

Many major companies are already working with NVIDIA on this effort. Partners include Adobe, Atlassian, SAP, Salesforce, ServiceNow, Siemens, Cisco, CrowdStrike, Red Hat, Box, Cadence, Cohesity, Dassault Systèmes, IQVIA, and Synopsys. For example, Salesforce is developing a system where employees can manage AI agents through Slack, pulling data from both cloud and on-premises sources, all powered by NVIDIA infrastructure.

Atlassian is integrating the toolkit into its Rovo AI strategy within Jira and Confluence platforms. ServiceNow is building an “Autonomous Workforce of AI Specialists” using NVIDIA AI-Q. Siemens has launched the Fuse EDA AI Agent, which uses NVIDIA Nemotron to automate workflows from design to manufacturing. IQVIA has already deployed over 150 AI agents across internal teams and client projects, including many of the top pharmaceutical companies.

This growing list of users shows how NVIDIA’s new tools are starting to make a real impact. The focus is on creating a safer, more controlled environment for enterprise AI agents, enabling companies to unlock new efficiencies without sacrificing security or oversight.

Overall, NVIDIA is positioning itself as a key provider of the infrastructure needed for trustworthy, scalable AI deployment in the enterprise world. With tools like OpenShell and AI-Q, the company is aiming to make autonomous AI agents not just powerful, but safe and cost-effective for widespread enterprise use.

Inspired by

• https://www.artificialintelligence-news.com/news/nvidia-agent-toolkit-enterprise-ai-agents/

Sources

• nvidia.com
• ai-expo.net
• techexevent.com
• techforge.pub
• techforge.pub