The self-hosted platform is available for internal pilots, with Red Hat and NVIDIA helping develop it after OpenAI donated the project to the OpenClaw Foundation.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
Primary source: [X](https://x.com/openclaw/status/2105023990607786313)
Why it matters #
OpenClaw is trying to make enterprise governance part of the open-source agent stack instead of selling it as a separate control layer. Its internal-pilot status and model-provider requirements make the maturity and operating costs as important as the free license.
The OpenClaw Foundation on September 29th released OpenClaw Enterprise, an open-source control plane for deploying and governing persistent AI agents on an organization's own infrastructure. The foundation says it will always be free for organizations to use, and says the project is available now for internal pilot workloads.
OpenClaw's creator, Peter Steinberger (@steipete), built his earlier developer-tools company, PSPDFKit, as a bootstrapped business after graduating from TU Wien and teaching iOS and Mac development there. The university's account of his career describes a founder who moved from freelance iOS work to building a commercial PDF framework. OpenClaw Enterprise extends that builder's open-source project into a different kind of product: infrastructure intended to make agents manageable inside organizations that might otherwise block them.
In its announcement, OpenClaw says organizations have asked for stronger security, safety and governance before deploying agents broadly. The new platform adds multi-tenancy, security boundaries, lifecycle governance and auditability. It is designed to let administrators swap out the agent harness, model and sandbox for third-party or internal alternatives, rather than requiring a single vendor's complete stack.
That flexibility is central to the pitch. Persistent agents can retain state and take actions over time, which gives them a different operational footprint from a chatbot responding to a single prompt. For an IT department, the control plane is meant to provide a place to set boundaries and track activity while agents run inside company-managed environments. OpenClaw's own announcement says the aim is to let organizations deploy capable agents without stripping away their abilities.
OpenClaw Enterprise grew out of work at OpenAI, which donated the project to the OpenClaw Foundation. The foundation says it has since developed the platform with Red Hat and NVIDIA. The arrangement gives the project support from companies with interests in enterprise software, AI models and computing infrastructure, while placing stewardship under an independent nonprofit. The foundation describes itself as an independent 501(c)(3); OpenClaw's project README says donors do not own or direct the project.
The code is public in the OpenClaw Enterprise repository, and organizations can deploy it with Docker Compose for local development or Kubernetes on internal infrastructure. The repository describes an OpenClaw Control Plane for deploying and managing agents, with components for identities and role-based authorization, resource lifecycle management, and audit events. OpenClaw says the project is being developed in public ahead of a planned 1.0 release later in 2026.
The current setup also puts a boundary around the vendor-neutral promise. The repository's getting-started guide requires an OpenAI API key to deploy a first agent using its default model, while allowing users to select an override. That requirement does not make the control plane OpenAI-only, but it shows that model choice and the platform's deployment experience are separate questions. The software can be self-hosted and designed for model substitution; running an agent still requires access to a model provider.
The free offer removes a license charge for the control plane, not the cost of operating agents. Organizations still supply infrastructure and model access, and must assess whether the software's early-stage controls meet their own security requirements. OpenClaw itself frames the present release as suitable for internal pilots, not as a finished 1.0 platform. Its bet is that opening the code before that milestone will bring outside users and contributors into the work of making the controls useful in production.
For Steinberger, whose earlier company sold developer infrastructure to businesses, this is another move from a tool built by a technical founder toward software other organizations can adopt. The difference is the ownership model: the foundation is offering the enterprise control layer without a paid tier, while the named corporate collaborators include the companies supplying models, infrastructure and enterprise software. Whether that combination draws cautious IT teams will depend on the project's maturity and the governance organizations can verify in practice, not on the word "enterprise" in its name.