“Think of it as Kubernetes for agents”: OpenClaw lands in the enterprise with Nvidia and Red Hat on board
The New Stack Paul Sawers
OpenClaw is turning its viral agent project into an enterprise platform. Nvidia and Red Hat are in, but the software is still pre-1.0 and half-built.
Based on reporting by The New Stack, Paul Sawers — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenClaw has gone from weekend curiosity to enterprise bet in less than a year. The self-hosted AI agent, created by Austrian developer Peter Steinberger in late 2025, hit more than 100,000 GitHub stars by February, when OpenAI hired him. Since then, the project has tried to shed the feel of a clever demo and become something companies might actually trust with real systems.
That push now has a name: OpenClaw Enterprise, or OCE. Kevin Lin, a member of technical staff at OpenAI who is leading the effort, says the basic problem is that agents are too hard for many companies to control centrally, so IT departments default to banning them. OCE is meant to answer that with a vendor-neutral, open-source control plane for deploying agents, isolating them, setting permissions, managing credentials, and keeping records of changes.
The architecture is already taking shape. OpenClaw says the API, console, persistent worker, PostgreSQL backend, and Kubernetes packaging are in place, while external gateway admission, workload authentication back to the control plane, and some model-authentication paths are still unfinished. The code is already on GitHub, and the current advice is blunt: this is for internal pilot projects only, not a finished enterprise product.
The Kubernetes analogy is doing a lot of work here, and OpenClaw is leaning into it hard. The official line is “Kubernetes for agents,” and the deployment model follows that script: the full local environment runs the control plane, PostgreSQL, and agent workloads inside Kubernetes, and companies can install it into Kubernetes infrastructure they already run. There’s also a Docker or Podman Compose option, but only for a control-plane preview that cannot deploy agents through OCC.
OpenClaw’s backing is unusually crowded for something this young. The Foundation counts OpenAI, Nvidia, Red Hat, and GitHub among its sponsors, and Red Hat and Nvidia have contributed resources. OpenAI and Red Hat are already testing OCE internally. Even so, OpenClaw says security is still the main focus, with isolation between trusted and untrusted workloads, sandboxing, granular permissions, and LLM-assisted review all part of the plan. The message is clear: the future is enterprise, but the plumbing is still being laid.
My take — AI-written commentary, not fact-checked reporting
This is the right move, because agents without a control plane are just expensive chaos with a logo. The real story isn’t the Kubernetes comparison; it’s how quickly the industry has moved from “wow, look what it can do” to “who gets access, who audits it, and who cleans up when it goes sideways.”
Read more about this at: The New Stack
