Back to blog

Why OpenClaw Enterprise?

Why companies should use OpenClaw Enterprise to manage agents across teams on infrastructure they control.

Miguel Branco
Miguel Branco
Founder & CEO, RAW Labs
6 min read

The outlined OpenClaw Enterprise mascot on the roof of an office building.

Here’s how agent deployments often start. Someone on the engineering team sets up their first agent using OpenClaw. They give it a repository and a test environment, and soon the agent is investigating failed builds autonomously: reading the logs, reproducing the failure, preparing a fix, running the tests. A colleague can step in halfway through, correct an assumption and send it after a different explanation.

Then the sales team wants one as well. That agent now needs commercial records instead of code repositories. It should not be able to use any of the engineering agent’s credentials. And because it is for the sales team, the person who configures is not the same person to deploy or manage it. Someone now has to decide where its conversations are stored, because they contain customer information. And when engineering changes their setup next month, the sales team’s agent should not change underneath.

This is the work of running agents as a company, rather than as a personal assistant or for a single small team. Without a shared solution, every team builds its own solution, own credential handling, own update procedure, its own admin interface.

OpenClaw Enterprise (OCE) is the shared answer. It is an open-source control plane that manages a fleet of OpenClaw agents across teams, on infrastructure you run. Think of it as “Kubernetes but for running agents”.

Why OpenClaw

OpenClaw is the most popular open-source agent. It started as one developer’s side project and became the fastest-growing project in GitHub history. It gave birth to the personal agent, behaving more like a colleague than a chatbot. You message it in Slack, Teams or WhatsApp. It does the work with real tools, remembers what it learned, and can carry a long task through on its own, within the limits you give it.

The early versions were built for one person on one laptop, and security teams were right to ban them. That is no longer the product. OpenClaw now belongs to an independent non-profit foundation. And now it also has an Enterprise edition backed by OpenAI, Red Hat and NVIDIA among others. Its security has been rebuilt from the ground up, and it has passed an independent security audit. In fact, OpenClaw is widely used at OpenAI to manage important services.

We also opted for building agents on OpenClaw for several reasons:

  • It is capable. An agent that does real work, rather than answering questions, is hard to build well. It has to plan, use tools, recover from its own mistakes and keep its context across a long task. OpenClaw does this with the frontier model of your choice, and its community tests it on real work every day.
  • It is widely adopted. As the fastest-growing project in GitHub history, it is the agent with the largest community, ecosystem of integrations and body of shared experience behind it.
  • It gets better every day. Thanks to a massive community of contributors, every new model, integration and fix that any user contributes reaches your deployment. Our effort, and yours, goes into what is specific to your business.
  • Its security work is public. The findings and the fixes are there for your security team to read.
  • It is widely used in production. OpenAI, for example, reports a fleet of over 500 OpenClaw agents doing important work around the clock.

What OpenClaw Enterprise (OCE) adds

OCE is built by people who run agents and infrastructure at scale. It began at OpenAI, with internal pilots at OpenAI and Red Hat, and was donated to the OpenClaw Foundation, where it is developed in collaboration with Red Hat, NVIDIA and other contributors. OpenClaw itself has maintainers from Microsoft, NVIDIA, OpenAI, Red Hat and Tencent.

OCE answers the questions that come up as soon as you run agents in production:

  • One control plane for every agent. Each team gets a Namespace. Each agent runs as its own runtime, with its own gateway and version history, so agents never share a conversation environment. You manage all of them centrally, from one console, API and CLI.
  • Access is granted, never inherited. Reading, editing and deploying an agent are authorized separately, per resource. Belonging to a Namespace grants nothing by itself, and an agent does not get its creator’s access. Secrets are bound to an agent explicitly, and their values never appear in responses, revisions or audit records.
  • Every change is deliberate and recorded. A deployment creates an immutable revision; editing a draft leaves running agents alone. Each change is written to the audit log, and if the audit record cannot be written, the change does not happen.
  • Limits are enforced outside the model. With dedicated execution, the gateway that handles messages and the harness that runs model turns and tools are in separate pods, with separate identities, storage and network policy.
  • Nothing is hard-wired. Compute, identity, secrets and channels are pluggable Drivers, and the model, harness and sandbox can be replaced.

You own it

An agent’s working material combines customer records, technical findings and internal discussion. Together they say more about a company than any one of the systems they came from.

OCE is open source and self-hostable. You decide where it runs and who administers it, and you can read the code. It is MIT-licensed and free to use.

The same applies to leaving. OpenClaw works with several model providers and with local endpoints, so the tools, procedures and evaluations you build survive a change of model (though you will want to re-test). Because the platform is open as well, you can operate it yourself, modify it, or get support from someone other than us.

If you are already on a hyperscaler

OCE can run in any managed Kubernetes service, so you can use what is already there: your cloud provider’s identity provider, its key vault, its monitoring and its hosted models.

What changes is who owns the layer that manages the agents. A hyperscaler’s agent service is part of that cloud and stays there. OCE is separate, open-source software, so the same agents, permissions and revisions can run in a Swiss or EU cloud, or on your own hardware with a local model, for the data that requires it. Sovereignty becomes a choice per workload, not one decision for the whole company.

What to know before you start

  • It is pre-1.0. OCE is still evolving rapidly, so we recommend starting with pilots, on pinned and tested releases, which is how we run it for customers.
  • The audit log covers management, not execution. It records resource changes, authorization denials and deployments. Tracing of what an agent did during a run is planned, not shipped, so we set up tracing alongside it.
  • Self-hosting covers the platform, not what you send out of it. An external model sees whatever you send it. Confidential data needs a local or sovereign model, chosen per workload.
  • Someone has to operate it. The software is free; running a platform is not. If you want a cloud provider to run everything, the managed agent services from AWS and Microsoft are good. If you want to own the platform without staffing it, that is what we do.

Where we come in

RAW Labs sells support to OCE as Enterprise OpenClaw. That covers supported combinations of runtime, integrations and configuration, tested updates, compatibility work and security triage. We also help with deployment and our services team can help you build and connect agents to your business systems. For more information, talk to us.

Stay up to date with AI.

AI news and insights from RAW Labs, covering the technology, the business decisions, and what it takes to put AI to work.

Unsubscribe anytime. Privacy policy