OpenClaw Enterprise is here: what an always-on AI agent needs from your hardware

Picture an AI assistant that does not wait for you to open a tab. It sits on a machine in the corner, reads your messages, drafts replies, handles files and works on tasks while you sleep. That is the idea behind OpenClaw, the open-source agent, and on 29 September 2026 the project announced OpenClaw Enterprise, a free platform for running such agents with more control. Here is what the news means, and what it means for the machine you would run one on.

What is OpenClaw Enterprise?

A free, open-source platform for running persistent AI agents with stronger security and central oversight, released under the MIT licence. It gives organisations one control layer to deploy agents, set rules and track what they do. It sandboxes agents, applies fine-grained permissions, keeps audit records across an agent's life, and lets teams swap in their own model or tools. The project began at OpenAI and was donated to the OpenClaw Foundation, and Red Hat and NVIDIA are collaborating on it.

Is it ready to use?

Not for everything yet. It is still under development, with a 1.0 release planned for later in 2026, and its makers recommend it for internal pilots rather than broad production. It deploys on Kubernetes, and the software is free but you still pay for the computers and the AI models behind it. The reporting also recalls earlier OpenClaw problems, such as data leaks through indirect prompt injection, which concerned earlier versions, so check isolation and permissions before trusting it with anything important.

What machine do you need at home?

Less than you might think, if the agent uses a cloud AI model. A setup guide from Astropad, a company that sells remote-desktop software, says any Apple Silicon Mac can run OpenClaw and recommends a base Mac mini with 16GB of memory and 256GB of storage. The Mac mini with 16GB and 256GB matches that specification. Most of the thinking happens on the model provider's servers, so the Mac needs few resources, and it is small, quiet and cheap to leave on all day.

What if you want to run the model yourself?

Then memory becomes the constraint. The same guide suggests 32GB only if you plan to run local models later. A larger machine such as the Mac mini M5 Pro with 24GB gives room for bigger local models, and for the largest ones a mini PC with 128GB of shared memory, like the GMKtec EVO-X2, can use roughly 96 to 110GB of it for AI. Running locally saves per-use fees but needs the hardware to carry it.

What does it cost to run with a cloud model?

The guide estimates roughly 5 to 20 US dollars a month for light use, 20 to 80 for moderate use, and over 100 for heavy use. The machine is a one-time cost and the model fees are the running cost, which is why some people eventually move to local models.

How do you keep it safe?

• Use a dedicated machine, separate from your main computer, so the agent cannot reach your passwords and private files.

• Pair it so that only you can send it commands.

• Remember the trade-off in always-on settings: the guide turns off disk encryption so the Mac can restart unattended, which makes a dedicated machine even more important.

• Treat anything the agent reads from the web as untrusted.

Which should you choose?

• Trying an always-on agent with a cloud model: a 16GB Mac mini, kept separate from your daily computer.

• Want room for bigger local models: more memory, such as the 24GB Mac mini.

• Want to run very large models without per-use fees: a 128GB mini PC.