Cubicles
Cubicles

What is OpenClaw? A plain-English guide to the open-source AI assistant

The open-source assistant that answers on your chat apps and does the work on a computer of its own. What it is, where it came from, and what you need to run it.

Blog ·

OpenClaw is a free, open-source AI assistant that runs on a computer you control and talks to you through the chat apps you already use. You message it on Telegram, WhatsApp or Discord. It does the work with a browser, a terminal and files on its computer, then messages you back.

That's the difference from a chatbot. ChatGPT or Claude in a browser tab answers questions. OpenClaw takes actions: it opens websites, fills in forms, runs scripts, saves files and checks back on a schedule, even when you're not talking to it.

OpenClaw at a glance

OpenClaw
What it isA personal AI assistant that takes actions, not just a chat window
LicenceMIT, free and open source
Written inTypeScript, running on Node.js
Runs onLinux and macOS, on any computer that stays on
Talks to you onTelegram, Discord, WhatsApp, Signal, Slack and more
Uses the model fromAnthropic, OpenAI, Google, OpenRouter, DeepSeek and others, with your own API key
Looked after byThe OpenClaw Foundation, a nonprofit, since mid-2026

OpenClaw changes quickly. This reflects the project as of November 2026: check the official docs for the current details.

Where OpenClaw came from

Austrian developer Peter Steinberger started the project in late 2025 as Clawdbot, a pun on Claude. It spread fast. In January 2026 Anthropic raised a trademark concern about the name, and it became Moltbot. A few days later it was renamed again, to OpenClaw, because Moltbot was hard to say.

In February 2026 Steinberger joined OpenAI. OpenAI hired him; it didn't buy the project. OpenClaw stayed open source, and in July 2026 it moved to the OpenClaw Foundation, a nonprofit set up to keep it independent, under the same MIT licence.

So if you see guides for Clawdbot or Moltbot, they're about the same software, just older versions.

How OpenClaw works

OpenClaw is a few pieces working together:

PieceWhat it does
The gatewayAn always-running process that connects your chat apps, your devices and one or more agents. It's what makes OpenClaw reachable from your phone.
The modelThe AI that does the thinking. OpenClaw doesn't include one: you connect a provider with your own API key.
ToolsA browser, a terminal and the file system, so the model can act rather than just reply.
The workspaceA folder of Markdown files holding the agent's instructions, notes and memory. You can read and edit them.
SkillsPackaged instructions, and sometimes code, that teach it a task or a service. Thousands are shared on the community registry, ClawHub.
AutomationCron jobs for exact times and a periodic heartbeat for regular check-ins, so it can work without being asked.

A typical exchange: you message “find me three quiet hotels near the conference venue under $200 and save the options”. The gateway passes that to the agent. The model decides to search, opens a few booking pages in the browser, reads them, writes a short comparison to a file and sends you the summary on Telegram.

What people use OpenClaw for

  • A morning brief. Calendar, weather, overnight email and news on a topic, waiting on your phone at 7am.
  • Inbox triage. Sorting, labelling and drafting replies for you to check. See how to let an AI agent manage your inbox.
  • Research. Comparing suppliers, products or prices across many sites and saving a table with sources.
  • Watching for changes. A price drop, a new job posting, a page that updates, checked on a schedule.
  • Small admin jobs. Filling in forms, renaming files, pulling numbers from one site into a spreadsheet.
  • Light coding. Scripts, small fixes and automations, though dedicated coding agents suit bigger work.

What it isn't good at (yet)

  • Long, vague tasks. It does best with a clear goal and a clear finish. “Grow my business” goes nowhere; “list ten local competitors and their prices” works.
  • Getting everything right. It can misread a page or invent a detail. Ask for sources and check anything that matters.
  • Sites that block automation. Some sites challenge anything that looks like a bot, and it may need you to sign in or pass a check.
  • Telling good instructions from bad. A web page or email can contain instructions aimed at it. More on that below.

What OpenClaw costs

The software is free. You pay for two things: the model, billed by your AI provider for what the agent reads and writes, and the computer it runs on. The model bill depends heavily on how much you use it and which model you pick. Our guide to what an AI agent costs to run breaks both down, and choosing a model for your agent covers the trade-offs.

Where to run OpenClaw

OpenClaw needs a computer that stays on. On your laptop it stops when the lid closes, and it runs next to your personal files and logins. The usual options are a spare machine at home, a Mac mini, a server you rent, or a cloud computer built for agents.

Each has trade-offs, covered in where should your AI agent run?. If you're weighing hardware, read do you need a Mac mini for OpenClaw? first.

Is OpenClaw safe?

It's as safe as the way you run it. The risks are real and well documented:

  • Prompt injection. Text in a web page or email can try to give the agent orders. No AI agent fully solves this yet.
  • Untrusted skills. Skills are software written by strangers, and researchers found hundreds of malicious ones on public registries in early 2026.
  • Too much access. An agent signed in to your main email, bank and cloud storage can do a lot of damage with one bad instruction.

The fixes are practical: run it on its own computer, not yours; give it its own accounts with only the access the job needs; read a skill before you install it; and keep payments and anything irreversible behind your approval. The AI agent security checklist walks through each step.

How to get started

Option 1: install it yourself

Install Node.js on a Linux or macOS machine, follow the setup in the OpenClaw docs, connect a model and a chat app, then set the machine up to stay on and restart the gateway after a reboot. Expect an evening if you're comfortable in a terminal.

Option 2: use a computer where it's already installed

On Cubicles, OpenClaw comes installed on every cloud computer, and the New agent form sets it up for you:

  1. Sign up. Your first computer is free, with no card needed.
  2. Choose New agent and pick OpenClaw.
  3. Choose a model provider and model, and paste your API key.
  4. Pick Telegram, Discord, WhatsApp or Signal. It answers only the account you name.
  5. Choose Continue to its computer and create it. OpenClaw is set up there a few minutes later.
  6. Send it a first small task, then open the computer in your browser to watch it work.

Each computer is a full Ubuntu desktop with 4 vCPU, 8 GB of RAM and 64 GB of NVMe storage. It keeps running when you close the tab. The full walkthrough is in OpenClaw hosting. For the chat side in detail, see how to chat with your AI agent on Telegram, Discord, WhatsApp or Signal.

OpenClaw and the alternatives

  • Hermes Agent from Nous Research is the closest alternative, built around memory and skills it writes itself. See OpenClaw vs Hermes Agent.
  • Claude Code, Codex and OpenCode are coding agents first. See Claude Code vs Codex vs OpenCode.
  • Hosted agents in ChatGPT and other apps need no setup, but run on the provider's computers, with less control over what they can reach and keep.

Frequently asked questions

Is OpenClaw free?

Yes. OpenClaw is open source under the MIT licence. You pay your AI provider for the model it uses, and for the computer it runs on.

Is OpenClaw the same as Clawdbot and Moltbot?

Yes. The project started as Clawdbot in late 2025, was renamed Moltbot in January 2026 after a trademark concern, and became OpenClaw a few days later.

Who makes OpenClaw?

Peter Steinberger created it. Since mid-2026 it has been looked after by the OpenClaw Foundation, a nonprofit, with a community of contributors.

Does OpenClaw need a Mac?

No. It runs on Linux too. A Mac is only needed for Apple-only integrations such as iMessage.

What can OpenClaw do?

It can browse websites, fill in forms, run commands, read and write files, and run jobs on a schedule, then report back on your chat apps. It works best on clear, well-scoped tasks.

Is OpenClaw safe to use?

It can be, if you run it on its own computer, give it only the accounts its job needs, check skills before installing them and keep payments behind your approval. Prompt injection remains an open problem for every AI agent.

Give your agent a computer