Enterprise adoption of autonomous AI agents has stalled less on capability than on control — who governs what an agent can do, and what happens when it acts on sensitive data.
NVIDIA announced the NVIDIA Agent Toolkit at GTC 2026 in San Jose on March 16, positioning it as an open-source software stack that gives enterprises and developers the infrastructure to build and deploy autonomous agents with enforced security and privacy constraints.
At the center of the toolkit is NVIDIA OpenShell, an open-source runtime that applies policy-based guardrails to individual agents — referred to internally as “claws.” Jensen Huang described the moment at GTC: “Claude Code and OpenClaw have sparked the agent inflexion point – extending AI beyond generation and reasoning into action. Employees will be supercharged by teams of frontier and custom-built agents they deploy and manage.” NVIDIA is building OpenShell compatibility with Cisco, CrowdStrike, Google, Microsoft Security, and TrendAI, integrating the runtime’s controls directly into those vendors’ existing security tools rather than asking enterprises to work around them.
The second major component is NVIDIA AI-Q, an agentic search blueprint built with LangChain that uses a hybrid model architecture: frontier models handle orchestration while NVIDIA‘s open Nemotron models perform research-intensive tasks. According to the announcement, this split reduces query costs by more than 50% while achieving top scores on the DeepResearch Bench and DeepResearch Bench II leaderboards.
That cost figure carries specific weight for enterprise buyers who have watched AI spending spike unexpectedly once pilots move to production.
The partner list for the toolkit spans Adobe, Atlassian, SAP, Salesforce, ServiceNow, Siemens, Red Hat, Box, Cadence, Cohesity, Dassault Systèmes, IQVIA, and Synopsys, among others. Each integration points to a different deployment pattern: Salesforce is building a reference architecture where employees use Slack as the orchestration layer for its Agentforce agents, pulling data from both on-premises and cloud environments. Atlassian is folding the toolkit into its Rovo AI work across Jira and Confluence. ServiceNow‘s “Autonomous Workforce of AI Specialists” runs on the toolkit with NVIDIA AI-Q underneath. Siemens launched the Fuse EDA AI Agent, using Nemotron to autonomously orchestrate workflows across its electronic design automation portfolio from design conception through manufacturing sign-off.
IQVIA provides the most concrete deployment figure in the announcement: the company says it has already rolled out more than 150 agents across internal teams and client environments, with coverage reaching 19 of the top 20 pharmaceutical companies.
Availability and Infrastructure
What NVIDIA is assembling — the Agent Toolkit, OpenShell, Nemotron models, and AI-Q — amounts to a software infrastructure layer designed to sit beneath enterprise software deployments. The toolkit is available now on build.nvidia.com, with support across AWS, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure.
Photo by Pixabay
This article is a curated summary based on third-party sources. Source: Read the original article
