The OpenClaw Foundation has released OpenClaw Enterprise, an open-source control plane for running and governing many long-lived AI agents inside a company, VentureBeat reported Tuesday. The project began inside OpenAI, counts Red Hat as a founding foundation member and uses a sandbox runtime contributed by Nvidia.

What Happened

OpenClaw Enterprise, or OCE, sits a layer above individual agents. Rather than defining how an agent behaves, it handles the infrastructure around it: multi-tenant administration, fine-grained permissions, workload isolation and sandboxing, auditing, and reviews of agent actions by a language model. The project's GitHub repository describes it, in VentureBeat's account, as something like Kubernetes for agents. Models, agent harnesses and sandboxes are designed to be replaceable.

The code is released under the MIT license and is free to use. Companies pay only for the compute, models, storage and infrastructure they run it on. It can be self-hosted with Docker Compose for local development or on Kubernetes for production.

RelatedClaude Code Desktop Preview Flaw Could Have Exposed Developers’ Cloud Keys, Researchers Say

OpenAI built the system for its own use before donating it to the foundation, and VentureBeat said both OpenAI and Red Hat are piloting it internally. Red Hat is working on OpenShift isolation that places agents in separate namespaces and routes credentials through proxies, treating agent processes as untrusted. Nvidia contributed OpenShell, an open-source runtime that puts agents in isolated environments with deny-by-default permissions and audit trails.

The foundation currently recommends OCE for pilot workloads only. It plans a 1.0 release later in 2026, along with a reference architecture describing workload boundaries and how permissions integrate with existing enterprise systems. OpenAI technical staff member RJ Marsan described an internal agent's ability to trace problems and publish fixes as "kinda game changing," according to VentureBeat.

Why It Matters

Persistent agents that hold credentials, run for days and act across internal systems create a different risk profile from a chatbot answering questions. Security teams need to know which agent did what, under whose authority, with which data, and how to shut it down. OCE packages those controls into one open layer that does not tie a company to a single model vendor.

The backing matters as much as the code. An OpenAI origin, a Red Hat role in the foundation and an Nvidia runtime give the project credibility with enterprise buyers who might otherwise wait for a commercial platform. The permissive license also lets cloud providers and software vendors build products on top of it.

Competition

The agent-governance category is filling up with venture-backed startups. VentureBeat pointed to Runlayer, which raised a $30 million Series A in June from Felicis and Khosla Ventures, bringing its total funding to $42 million. A free, foundation-run alternative with major-vendor support puts pressure on paid tools to show clear advantages in integration, support or compliance features.

What to Watch

The 1.0 release and reference architecture will show whether OCE is ready for production use, and whether it integrates cleanly with existing identity and access management systems. Adoption outside OpenAI and Red Hat, contributions from other vendors, and whether cloud providers offer managed versions will indicate if it becomes a common layer or remains one option among many.