OpenClaw. Built for enterprise.
OpenClaw is the agent, already built. OpenClaw Enterprise runs it, and every agent after it, with the controls your security team asks for.
Who runs OpenClaw, and who builds on it
-
OpenAI
Runs it
Runs OpenClaw agents internally with full access to its codebases and plugins. Its engineers describe an agent that triages channels and fixes broken builds.
OpenClaw Foundation, September 2026 -
Microsoft
Runs it
Employees used an early build of Scout, now Autopilot, before its launch. Microsoft says the agent is powered by OpenClaw and contributes upstream.
Microsoft, June 2026 -
NVIDIA
Builds on it
NemoClaw, its one-command stack for secure always-on agents, is built on OpenClaw's MIT-licensed codebase and installs OpenClaw with its Nemotron models.
NVIDIA, April 2026 -
Cisco
Runs it
Early adopters across the company, called together by its EVP of Operations to share how they use OpenClaw at work.
Cisco, March 2026 -
Red Hat
Builds on it
Its OpenShift AI platform serves the models and runs the agents for OpenClaw workloads. An OpenClaw maintainer leads its upstream team.
Red Hat, September 2026 -
Tencent
Builds on it
ClawPro, its enterprise platform for OpenClaw, used in beta by more than 200 organizations before launch.
The Next Web, April 2026 -
University of Michigan
Runs it
Founded the Institute for Agentic Computing as the academic home for OpenClaw, where its researchers apply agents to science and engineering.
OpenClaw Foundation, July 2026 -
Xiaomi
Builds on it
MiClaw, an OpenClaw agent built into the operating system of its phones.
Yicai Global, March 2026 -
ByteDance
Builds on it
ArkClaw, a hosted OpenClaw on its Volcano Engine cloud, used from the browser.
Yicai Global, March 2026
A capable agent, on day one.
Memory, safeguards, tools and retesting on every new model take a long time to build. OpenClaw has all of it, kept up by its Foundation’s team and maintainers across the industry.
Your own build
Time spent
- MemoryBuilt
- ToolsBuilt
- Learned skillsBuilt
- ApprovalsStill to build
- SandboxStill to build
- FailoverStill to build
- SchedulingStill to build
- Sub-agentsStill to build
- Model testingStill to build
OpenClaw
Day one
- MemoryBuilt in
- ToolsBuilt in
- Learned skillsBuilt in
- ApprovalsBuilt in
- SandboxBuilt in
- FailoverBuilt in
- SchedulingBuilt in
- Sub-agentsBuilt in
- Model testingBuilt in
Secure, and your team can check it.
One set of controls for every team’s agents. Each agent runs with its own identity and only the access granted to it, in a space of its own. Every change is on record, and the code behind every control is open.
OpenClaw Enterprise
-
Identity
Each agent has its own.
-
Permissions
Who may view, edit or deploy.
-
Boundaries
A namespace for each team.
-
The record
Every deployment, every refusal.
Extensible, even to your own systems.
Dozens of chat apps and over a hundred plugins come ready, and your engineers can plug in your own systems.
- Slack
- Teams
- Discord
- Telegram
- Signal
- GitHub
- Notion
- Jira
- Salesforce
- HubSpot
- Zendesk
- Odoo
- Your system
Any model, frontier or open-weight.
Each agent picks its model: OpenAI’s or Anthropic’s frontier models, or an open-weight model behind an OpenAI-compatible endpoint. When a better one comes out, switch it; the agent stays the same.
- Memory
- Tools
- Permissions
OpenAI: Frontier models
Sovereign, down to the model.
Open-weight models run on your own servers, so no prompt, document or answer goes to a model provider. We set up memory search and update checks to match.
-
Your servers run the model
-
Your network carries prompts and documents
-
Your storage keeps memory and logs
No vendor lock-in.
Both projects are open source under the MIT licence and run on Kubernetes in any cloud or your own data centre. Change cloud, model vendor or partner, and your agents come with you.
OpenClaw or Codex, whichever fits the task.
Codex, OpenAI’s coding agent, is made for work in code and the terminal. OpenClaw remembers, learns skills and works on a schedule, on any model. Each agent runs on the one that suits its work.
Scales from one agent to every team.
Every agent runs as a workload of its own on Kubernetes, in its team’s space and within the team’s resource limits. Add agents, and teams, under the same rules: the security review doesn’t start over.
Your infrastructure. Or ours.
The platform runs on a Kubernetes cluster you prepare, or on one we provide. Your team operates it, or RAW Labs does.
Three small models in line drawings, one for each place, each a row of three server cabinets: one in front of an office building, one in front of a cloud, one on a round base that carries RAW Labs’ logo. In every row the middle cabinet is the platform’s, drawn in red and wearing OpenClaw Enterprise’s mark.
-
On premises
Your own data centre
-
Your cloud
The cloud you already use
-
Our infrastructure
Provided and run by RAW Labs
Linux nodes, AMD64 or ARM64. Images from public registries or your own mirror.
Tested as one. Shipped as one.
RAW Labs pins one exact commit of each project and releases the two as one.
OpenClaw OpenClaw Enterprise
OpenClaw
OpenClaw Enterprise
- Built for AMD64 and ARM64
- Tests passed
- Installed on a clean cluster
- Never changed once published
Fair questions. Plain answers.
-
What is Enterprise OpenClaw by RAW Labs?
Enterprise OpenClaw is a maintained, supported bundle built on two open-source projects. OpenClaw is the runtime for persistent agents: each has a workspace, memory, tools, application connections and a schedule. OpenClaw Enterprise includes OpenClaw and provides company-wide controls: identity, permissions, namespaces per team, deployment revisions and an audit log. RAW Labs helps you deploy, operate and maintain it, with expert support.
-
Why OpenClaw?
Because it is the open-source agent the community chose: a runtime anyone can read, run and improve, an independent foundation behind it, and a control plane led by OpenAI and built in the open with Red Hat and NVIDIA. We would rather build on that than on a black box, and we’re committed to contributing our improvements upstream.
-
Do we need OpenClaw Enterprise, or is OpenClaw enough?
OpenClaw runs the agents. OCE manages them across teams, with separate identities, permissions and deployment histories. RAW Labs ships and supports the two together.
-
What does RAW Labs add?
Releases of the two projects pinned to exact versions, built for AMD64 and ARM64 and checked in a reference installation, with fixes shipped as new releases. On top of them, our own enterprise connectors to business applications, agent configurations and other custom plugins, reviewed security settings and sandbox profiles, and an upgrade and recovery procedure. Support comes with an agreement: expert help, ongoing maintenance, the versions we maintain, security fixes, response times and a team to call. Your subscription covers maintenance and support for the release you use.
-
Is it open source, or are we locked in?
OpenClaw, from the independent OpenClaw Foundation, and OpenClaw Enterprise are MIT-licensed open source, and RAW Labs’ own code in the bundle is licensed under Apache 2.0. The bundle builds OpenClaw Enterprise from RAW Labs’ public fork of it, naming every commit in each release, and we’re committed to contributing our improvements upstream so the wider community benefits. Your agents, their configuration and their state live in your installation.
-
Where does our data live?
On your own Kubernetes, on premises or in your cloud, whether your team or RAW Labs operates it; in RAW Labs’ infrastructure only when RAW Labs runs it. Model calls go to the providers you configure, application connections use accounts you authorise, and memory and conversation history keep to the retention policy agreed for the deployment.
-
Which applications and channels can agents use?
Hundreds of integrations supported by OpenClaw, plus our own enterprise business connectors, tested for each deployment and to each agent’s needs. Agents reach the business applications connected for your deployment, using accounts you authorise, and access can be limited per agent.
-
Which models can we use?
OpenAI and Anthropic models, NVIDIA-hosted models, and local or self-hosted models through compatible endpoints. Model usage is billed by your provider, separately from the subscription.
-
How do teams work on the same platform?
Teams work in separate namespaces, each with its own agents, configurations and credentials. Platform operators manage the infrastructure and access policies across them.
-
What can we trace when an agent changes?
Deployment revisions preserve configuration history. The audit log records platform changes and their actors; prompts and tool activity need separate logging. See the white paper .
-
What do we need to run it?
A Kubernetes cluster prepared as OpenClaw Enterprise’s production guide specifies: PostgreSQL, DNS and TLS, storage, separate control and runtime node pools, enforced network policies, cert-manager and Envoy Gateway, on Linux AMD64 or ARM64 nodes. A workstation with Python 3.12 or newer, Node 24 or newer, Bash, Helm 3, kubectl, jq, yq and openssl. A key for your model provider. Or let RAW Labs run it for you, on your infrastructure or on ours.
-
How do upgrades work?
Each release names the exact versions it pins, of the runtime, control plane, plugins and deployment, and its known limitations, tested as one combination, and packages OpenClaw Enterprise’s upgrade and recovery procedures. Release checks can include an upgrade from a selected earlier release; confirm the tested path and compatibility notes with your support contact first. An update redeploys the running agents together, and a Helm rollback doesn’t reverse database changes. Your team applies it, or RAW Labs does under managed operations.
-
Who operates it?
Your team on your Kubernetes, RAW Labs on your infrastructure, or RAW Labs as the host. In every case a named operator owns monitoring, updates, incidents, backups and access administration, agreed before handover.
-
What support is included?
Product support at the level agreed in the quote, business hours to start, with a named contact channel and response targets. It covers platform defects, security triage, compatibility and operating guidance; custom agents and workflows are maintained under a separate agreement.
-
Do you build the agents as well?
Yes. Our services team prepares the environment, deploys the bundle, connects your systems and delivers the first workflows. Customer solutions build agents and integrations with agreed permissions and approvals, and training and handover prepare your users, administrators and operator.
-
How is it priced?
An annual subscription for the agreed installations and agent capacity covers the maintained bundle, tested updates, standard integrations, documentation and product support. Implementation is a fixed project fee, managed operations and solution maintenance recur, and infrastructure, model usage and the connector service are billed separately.
-
How do we start?
With the work you have in mind and where it should run. An assessment finds the first agent, or contact us about the deployment itself and we scope it together: environment, integrations, first workflows, owners and acceptance criteria.