What OpenClaw is, and what you need to run it

OpenClaw is an open-source personal AI agent that runs on your own server. Instead of opening a vendor’s website to talk to an AI, you install the program on a machine of yours, and that machine receives your requests, acts on them and replies. To run it properly you need a VPS with root access, a recent version of Node.js and an API key from a model provider.

One warning up front: OpenClaw is a third-party project. MozOut does not sell it, install it or repair it. What we provide is the server it can live on. Where support stops is written in how far our support goes.

What it is, underneath

OpenClaw does not run AI models itself. It is a Node.js application that sends requests over the internet to cloud model APIs, such as those from OpenAI, Anthropic or Google, and uses the answers to act: read files, run commands, talk to you through a messaging channel. The model at the provider does the thinking; OpenClaw is what connects it to your machine and your accounts.

Only people who want to run the models on the server itself, for example with Ollama, need a powerful machine. That is a different road, covered in running models on your VPS with Ollama.

What you need, according to the project’s documentation

Requirement	What the documentation says	What it means for you
Node.js	Version 22.14 at minimum, 24 recommended	Many VPS images ship an older version, so you have to upgrade before or during installation.
CPU and memory	2 vCPU and 4 GB of RAM as a starting point for testing	A 1 GB VPS fails during installation for lack of memory. With browser automation or several channels, plan for more RAM.
Access	Root over SSH and a public IP	Only a VPS, Cloud or dedicated server gives you that. It does not work on shared hosting.
API key	From a model provider	The running cost is the cost of the model calls. See what it costs.

The figures above come from the project’s public documentation and can change. Before installing, always read the official OpenClaw documentation: that is where the install command and the current versions live. Do not copy them from an article like this one.

Before you decide

1	Check that you need an agent. If you only want to chat with an AI, an account on a chat service is enough and needs no server. See which AI tool suits which task.

2	Pick the server. Plans and specifications are on the VPS and Cloud pages. The system we suggest for beginners is covered in which operating system to choose.

3	Think about security first. The agent acts on your behalf, with access to files, commands and accounts. Read what it can reach and how to limit it before handing it a single key.

4	Install last. The general steps are in installing OpenClaw on a VPS.

An agent is not a chat. A chat only answers. An agent can act. Give it access to more than it needs and a model mistake, or a malicious instruction hidden in a text it reads, can do real damage. Start on a test VPS with nothing important on it.

On an unmanaged VPS, installing and maintaining the application is up to you. See who does what on a VPS.

Need a server with root access to try it on? Have a look at the VPS plans.

See VPS plans

In our plans

The documented starting point for OpenClaw is 2 vCPU and 4 GB of RAM, with root access over SSH. VPS1 (1 vCPU, 2 GB, 40 GB NVMe) and Cloud1 (1 vCPU, 2 GB, 25 GB) are below it, so for OpenClaw choose a step above, with at least 2 vCPU and 4 GB of RAM. The VPS and Cloud pages list the vCPU and RAM of every step, up to 32 vCPU and 64 GB on VPS6 and Cloud6. Our VPS plans use KVM virtualisation and give you full SSH access, which is what the project asks for. Shared hosting does not fit. If you only want to test, you can start on the smallest step that meets the figures and move up if browser automation or several channels need more memory.

Categories

Categories