OpenClaw: the free open-source AI assistant that lives in your chats

OpenClaw is an MIT-licensed assistant you install on your own computer or server. It answers inside WhatsApp, Telegram and other chat apps, then acts on the accounts you connect: inbox, email, calendar, flight check-in. There is no subscription and no hosted service.

How it works

You run the assistant on your own machine and connect it to the chat apps you already use. Messages arrive from those apps, the agent decides what to do, and the actions happen against accounts you have authorised, such as mail and calendar. Memory persists between conversations, so it does not restart from nothing each morning. Capabilities come from skills you install, and the model behind it is whichever provider you point it at.

Because it runs on your hardware, the traffic between you and the model is yours to route. There is no vendor account in the middle, and nothing to cancel.

What you need

• A machine with macOS, Linux or Windows, or a small VPS if you want it awake when your laptop is not.
• Node.js, which the official installer brings with it.
• A model provider key, or a local model if you would rather not send prompts anywhere.
• A chat account to talk to it through, and the mail or calendar account you want it to act on.

The current release is 2026.11.1 on npm, checked in November 2026.

What it does well

• Inbox work is the strongest case: triage, drafts, follow-ups, without opening a mail client.
• Calendar and travel tasks go through the same conversation, which is the whole point of a chat-native assistant.
• Memory makes it useful for recurring work rather than one-off questions.
• Running locally means the setup survives a provider outage and does not depend on a company staying interested.

Where it falls short

• You pay for the model, not the software. A busy assistant on a frontier model costs real money each month.
• Connecting mail and calendar is a permissions decision. Scope the account, and revoke when you stop.
• It is young and moving quickly. Interfaces changed repeatedly through 2026, and an update can move something you had learned.
• Everything is on you: updates, the machine, and the consequences of an agent that can send email on your behalf.

How to install OpenClaw

Option 1, one-click install: Hostinger deploys OpenClaw as a preconfigured Docker app on a managed server, so there is no command line and nothing to configure. It runs the assistant with multi-channel support and full control.

Option 2, install it yourself: use the official installer on openclaw.ai, which sets up Node.js and the dependencies, or install the npm package with npm install -g openclaw and connect your chat apps and model keys.

FAQ

Is OpenClaw free? The software is. MIT licensed, no paid tier, no hosted version. Model calls are billed by the provider you connect.

Do I need my own API keys? Yes. You choose the provider and pay it directly, which is also why it can run against a local model instead.

Can I run it on a server rather than my laptop? Yes, and that is the usual setup if you want it available all day. A small VPS is enough; the model runs elsewhere unless you host that too.

What can it actually do to my email? Read it, draft replies and send, depending on the permissions you grant. Start with a secondary account and a short leash.

How does it compare with ChatGPT? ChatGPT answers in a tab you open. OpenClaw answers in chats you already use, keeps memory across sessions and takes actions on connected accounts. It does not browse the web for you the way a hosted assistant does unless you add that skill.

What if I want it in a terminal instead? There are terminal-based agent tools for that job. This page is about the chat-app assistant, so start here if that is the surface you want.

Last checked: November 2026. Licence read from the repository, version from npm, and feature claims taken from the project's own documentation.