Your own OpenClaw AI agent, on your own server

OpenClaw is a ready-made server with a personal AI agent pre-installed on Ubuntu 24.04. Instead of a chatbot that only talks, you get an assistant that acts: you message it in Telegram and it answers questions, sorts files, pulls together reports and runs routine tasks on a schedule, then reports back in the same chat. Because the agent lives on a server rather than in an app on your phone, it can keep working in the background while your files and history stay on your own machine. Plans start at 9.35 EUR/mo with annual billing, the server is ready minutes after ordering, and there is nothing for you to install or configure from scratch.

Yes, OpenClaw was set up precisely so that non-technical people can use an AI agent. If you can chat in Telegram, you already know the whole interface: you write to the agent in plain language, like you would write to a person, and it answers in the same chat. Ordering takes minutes, and the panel then guides you through the two setup steps, adding your AI key and connecting your Telegram bot. There is no command line and no coding at any point unless you choose to dig in. The one thing you do need is an AI key from OpenAI, Anthropic or avots.ai, and if you get stuck, write to support and we will walk you through it.

An AI key is a personal access code you get from an AI provider, and it is what lets the agent think: OpenClaw handles your files, tasks and Telegram chat, while the actual model behind the answers runs at OpenAI (ChatGPT), Anthropic (Claude) or avots.ai. We do not provide or resell keys, and that is deliberate. Bringing your own means you choose the model, you see exactly what the AI usage costs, and you pay the provider directly instead of paying a markup hidden in our price. In the panel you paste the key once and can swap it or change the model later. If you are unsure which provider to start with, write to support and we will advise.

There are two plans. OpenClaw costs 11 EUR/mo and gives you 1 vCPU, 2 GB RAM and 20 GB NVMe storage; OpenClaw Pro costs 38 EUR/mo with 2 vCPU, 8 GB RAM and 120 GB NVMe. Both prices exclude 21% VAT. Paying for 12 months in advance cuts the price by 15%, which is how the base plan comes down to 9.35 EUR/mo. The server price covers everything on our side: the pre-installed agent, the control panel, the private machine and updates. What it does not cover is your AI key, which you buy from OpenAI, Anthropic or avots.ai and which that provider bills you directly.

The software is identical on both plans; the difference is the size of the machine underneath. The base plan has 1 vCPU, 2 GB RAM and 20 GB NVMe for 11 EUR/mo, while Pro has 2 vCPU, 8 GB RAM and 120 GB NVMe for 38 EUR/mo. In practice the disk matters most, because the agent keeps your files and chat history on the server, and documents add up over time. Extra RAM and CPU help when you hand off heavier tasks like building large reports. A sensible approach is to start on the base plan and move to Pro when you feel the limits; you do not have to decide everything on day one.

You choose the location when ordering. The first option is Riga, Latvia, in our own Tier 3+ data centre, running on our own network (AS58269), which means we control the hardware and connectivity end to end. The second is Amsterdam in the Netherlands, and Dubai is coming soon. Both current locations are in the European Union, so your files and chat history stay on a private single-tenant server under EU jurisdiction rather than in a big shared cloud. For most customers the practical difference is small, so pick the region closer to you or to the services your agent will work with; if in doubt, write to support.

Running tasks is the whole point of an agent, so the safety model is built around containment. Each OpenClaw runs on its own private single-tenant server: whatever the agent does happens inside a machine that belongs only to you, fully isolated from other customers. Only you can talk to it, because it answers through your own Telegram bot. On our side we keep the platform patched and take regular snapshots, so if the agent, or you, ever makes a mess, the server can be rolled back to a working state. The honest caveat: the agent acts on your own server and your own data, so treat it like any assistant and give it clear instructions.

Yes, and nothing is locked away. OpenClaw is fully open source, published at github.com/cloudhostinglv/openclaw-vm, so you can read the code, take a plain server, install the agent yourself and tune every setting to your liking: the model, the behaviour, the way it handles files and tasks. The ready-made product exists for everyone who does not want that work; ordering it gets you the same software pre-installed on Ubuntu 24.04 with a control panel on top. Keep in mind that with a self-install you also take on updates and backups yourself, while on our servers we handle patching and snapshots. Either way you are welcome to write to support with questions.

On OpenClaw Pro, yes. That plan runs a build with a real Chromium on the same machine, so the agent can open a page, sign in with credentials you give it, click through a portal, fill a form and report what it found. It is the difference between an assistant that talks about a website and one that actually uses it. Typical jobs are pulling numbers out of a supplier portal that has no API, checking an order status, or collecting the same report every morning. The base plan has no browser, because Chromium needs memory that a 2 GB machine does not have to spare.

Inside your own virtual machine, on your own IP address, in our data centre in the EU. Nothing is proxied through a third-party browser cloud, so the pages, the logins you hand over, the cookies and the results all stay on the server you rent. That matters for anything involving customer data or company accounts, because it keeps the whole flow inside one EU jurisdiction and one contract. You can watch what happened in the machine at any time, and you can wipe it by reinstalling the server.

No. The browser is restricted to the owner exactly like shell access: your messaging channel only answers your account, and even inside it the browser is off limits to any other sender. This also protects you from a subtler problem: a web page can contain text written to trick an AI agent into doing something, and keeping the browser owner-only limits who can steer it there in the first place. As with any powerful tool, give the agent only the credentials you are comfortable with it having.