OpenClaw: What It Is and How to Install and Set Up Your AI Agent

OpenClaw is an open-source personal AI agent. You install it on your computer or server, connect it to a model and a messenger, and then talk to it like an assistant, in Telegram or another chat app. Unlike a regular chatbot, it can act: work with files, open websites and run tasks on a schedule.

The project became popular fast in early 2026, and a large ecosystem of plugins, called skills, grew around it.

How OpenClaw differs from ChatGPT

• It lives with you. The agent runs on your machine or server, not in someone else's cloud. The model itself can still be a cloud one.
• It works through a messenger. You write to it in a familiar chat; it replies and sends results.
• It acts on its own. It can run tasks on a schedule, not just answer questions.
• It extends with skills. Install ready-made ones or write your own.

What you can hand to OpenClaw

• A morning summary: email, calendar, news on your topics.
• Watching websites and reporting changes.
• Sorting files in a folder: renaming, organising, extracting data.
• Drafting posts and texts.
• Running your scripts on command from the chat.

How to install OpenClaw

1. 01Machinea separate server
2. 02Installofficial repository
3. 03Modelkey in settings
4. 04Messengerallowlist only
5. 05Gatewayclosed to the outside

Exact commands change between versions, so follow the official instructions in the project repository. The general order is:

1. Decide where it will live. For always-on use, a separate server or computer is better than your work laptop.
2. Install per the instructions. Make sure you install from the official repository.
3. Connect a model. You need an API key for your chosen model. Keep it in settings, not in chat.
4. Connect a messenger. For example, your own Telegram bot. Restrict who can message the agent, or anyone can command it.
5. Check the gateway. This is the part that receives commands. It must not be exposed to the internet without protection.

Configuration: what to watch

• Who can give commands. Only you or your team, by allowlist.
• Which skills are installed. Each skill is code that gets access to your agent. Install only trusted ones.
• Which folders and services it can reach. The minimum needed for its tasks.
• Model spend. An agent working on a schedule makes requests constantly. Set limits.
• An action log. You should see what the agent did and when.

Security: the essentials

A personal agent with access to files and accounts is powerful, and it is easy to misconfigure. Typical problems: the gateway reachable from the internet, a skill from an unverified source, the agent given access to the whole computer. So:

• run the agent on a separate machine or in an isolated environment;
• block outside access to the gateway;
• do not give it access to banking or core work accounts;
• keep the project updated; vulnerabilities get fixed regularly.

OpenClaw or a custom agent system

OpenClaw is good as a personal assistant and for small tasks. If you need a team of agents with roles, rules, a decision log and failure protection, for example to run dozens of channels, it is better to build a system for the job. I use both approaches, depending on scale.

FAQ

Is OpenClaw free?

The project itself is open source and free. You pay for the model if you use a cloud one, and for a server if you do not run it on your own computer.

Can OpenClaw be installed on Windows?

It runs on the major operating systems, including Windows, but a Linux server is more convenient for always-on use. Check the requirements in the official instructions.

Which model should I connect to OpenClaw?

Any supported one, cloud or local. Use a strong model for complex tasks and a cheap one for simple, frequent ones.

What is the OpenClaw gateway?

It is the component through which the agent receives messages and commands from messengers and other channels. It must be password-protected and never exposed directly to the internet.