Nvidia reportedly plans enterprise AI agent platform NemoClaw

Nvidia is planning to launch an open-source platform for AI agents NemoClaw, according to people familiar with the plans.

The platform is intended to let enterprise software firms dispatch AI agents to perform tasks for their workforces and will be available regardless of whether customers’ products run on Nvidia chips, sources said.

Nvidia has pitched NemoClaw to companies including Salesforce, Cisco, Google, Adobe, and CrowdStrike, though it was unclear if any formal partnerships exist.

Nvidia plans to offer security and privacy tools for the project, and said that because the platform is open source, partners would likely receive early access in exchange for contributing.

The move follows interest in so-called “claws”, open-source agents that run locally and perform sequential tasks, including OpenClaw, which was later acquired by OpenAI.

🔗 Source: Wired

🧠 Food for thought

Implications, context, and why it matters.

NemoClaw builds on Nvidia’s existing open-source agent tooling

NemoClaw would not start from zero. Nvidia already maintains the open-source NeMo Agent Toolkit for enterprise software teams to build, profile, and optimize complex systems of AI agents 1.
NeMo Agent Toolkit aims to work across frameworks, with integrations for Google’s Agent Development Kit (ADK), Microsoft’s AutoGen, Semantic Kernel, and Amazon Web Services’ (AWS) Strands/AgentCore 2.
Public descriptions list safety and security tooling for workflows, including simulated scenario-based attacks plus evaluation runs 3.

Nvidia’s planned agent platform appears to be a strategic hedge beyond CUDA and its own chips

The plan for NemoClaw includes access even when products run on non-Nvidia chips, which gives it a hardware-agnostic stance for an Nvidia-led agent effort 4.
WIRED connects the work to courting enterprise software companies and adding extra security layers for agents. It also sits alongside an open-source push as leading AI labs build custom chips. Nvidia’s software stack still leans on CUDA, its programming platform for Nvidia GPUs 4.
A PayPal paper using Nvidia’s NeMo Framework says a fine-tuned Nemotron small language model cut overall agent latency by 49% and monthly GPU costs by 45% 5.
WIRED said Nvidia discussed NemoClaw with enterprise software companies, while adding that it was unclear whether those talks became formal partnerships 4.