What is OpenClaw? The open-source AI agent, explained
OpenClaw is free, open-source software that turns an AI model into a personal agent you message from WhatsApp or Telegram. What it does, how it works, and what it takes to run.
OpenClaw is free, open-source software that turns an AI model into a personal agent: something with its own memory, its own computer, and the ability to actually do things — manage your inbox and calendar, make and watch things for you, run scheduled jobs — all from the chat apps you already use, like WhatsApp and Telegram. You run it on a machine you control, and you plug in the AI model of your choice with your own key.
“The AI that really does things”
That’s OpenClaw’s own tagline, and it’s the right frame. A chatbot answers when asked and forgets you when the tab closes. An OpenClaw agent is persistent software running around the clock on a real computer: it remembers you between conversations, it can use its machine — files, a browser, scheduled tasks — and it can message you first when something needs attention. The project’s examples are everyday ones: organizing your inbox, sending emails, managing your calendar.
The second thing that defines it: you talk to it from apps you already have. OpenClaw connects to WhatsApp, Telegram, Discord, Slack, Signal, iMessage and more — there’s no new app to install and no proprietary interface to live in. Your agent is a contact in your messages.
Open source, and genuinely yours
OpenClaw is MIT-licensed open source, started by Peter Steinberger and now governed by the non-profit OpenClaw Foundation with a large community behind it — by mid-2026 it had grown into one of the most popular open-source projects in the world. Practically, the licensing means three things:
- The software is free. You pay for the AI model your agent thinks with (via your own API key) and for whatever computer it runs on — the honest cost breakdown is here.
- Your agent lives on hardware you control. Its memory, files, and personality are plain files on your machine — not rows in a vendor’s database.
- No lock-in on the brain. It works with the major model providers (Anthropic, OpenAI, Google, and many others), and you can switch models without losing your agent.
How it works, in one paragraph
At the center is a single always-on process called the Gateway. It holds your agent’s sessions and memory, connects outward to your chat apps, and routes every message between you, the agent, and the AI model. There’s also a web dashboard (the Control UI) for setup and management. Because the Gateway must be running for your agent to exist, the practical question every new OpenClaw user hits within days is: which always-on computer should it live on?
What it takes to run
Honestly: OpenClaw is developer-friendly rather than consumer-polished. Installing it is a few terminal commands, and connecting a chat channel means creating a bot token or scanning a QR code. None of it is hard for anyone comfortable with a terminal — but it is a piece of software you operate, with real security decisions that deserve adult attention, because an agent that can act on a computer is only as safe as the computer you give it.
That’s the honest trade at the heart of OpenClaw: it gives you an agent that is genuinely yours — owned software, your model key, your data — and in exchange, someone has to be its sysadmin. Doing that yourself is a fine weekend project; having it done for you is what managed hosting like Everpod exists for. Either way, the agent you end up with is the same OpenClaw.