OpenClaw

advancedspreading

An open-source, locally run personal AI agent that you instruct via a chat application and which independently handles tasks on your behalf.

In plain terms

OpenClaw is like a personal secretary who lives on your own computer and is reached by message. You write to it on WhatsApp or Telegram, and it not only replies but also does what you ask: sorts your email, manages your calendar, searches for files, opens web pages.

The standard chatbot mainly converses with you and works only when you write to it. OpenClaw, however, gets hands-on: it accesses your machine, accounts, and programs, and can work even when you are not paying attention. It can also initiate contact on its own, for example, summarizing your day in the morning.

OpenClaw itself is not a language model but the framework around one. The "brain" is provided by a model of your choice (for example Claude, GPT or a locally running model), and OpenClaw connects it to your messaging apps, its memory and its tools.

Free and open-source software, which is why a large community quickly formed around it. This same freedom is also its risk: whatever you entrust to it, it truly has access to.

How it works

Install it on your own machine or a server and specify which language model it uses for reasoning.
You connect it to your chat applications, so you can message it just like a friend.
When given a task, the model breaks it into steps, and OpenClaw executes them: running commands, controlling a browser, reading and writing files, and sending emails.
It records important information in simple text files on your computer, so it remembers your habits and previous conversations later.
New capabilities are added via so-called skills: downloadable guide and tool packages created and shared by the community.

An everyday example

In the morning on the bus, you message OpenClaw running on your home computer via Telegram to review your emails and book a dentist appointment. By the time you arrive, you receive a summary of important emails in a message, and the appointment is already in your calendar.

Who is affected

developerThey are the one who can install it safely, customize it, and extend it with their own skills, and who best understands how much access it gives to their computer. → Code

entrepreneur, executiveIt promises a cheap, always-available assistant, but if staff install it on company devices without permission, it poses a serious data security risk.

employeeIt can relieve you of routine tasks, but if you allow it access to corporate email or files, you remain responsible. → Work & HR

everyoneIt clearly shows what it will be like when AI not only answers but acts on your behalf.

A common misconception

MythOpenClaw is a new, smarter artificial intelligence, a rival to ChatGPT.

In realityIt is not a model but a framework: it connects an existing language model to your machine and your messaging tools. It is as smart as the model behind it and as dangerous as the access you grant it.

Limits and risks

Whatever access you grant (email, files, passwords, credit cards), it can also make mistakes: it may delete, send, or purchase something you did not intend.
A hidden instruction in a webpage or email can take control over it (this is called prompt injection) and turn it against you.
Malicious, data-stealing packages have already been found among community skills, so install only from trusted sources.
Configuration requires technical knowledge, and using the underlying model may cost money, even if the program itself is free.

Similar, but not the same

Chatbot – A chatbot primarily converses and responds with text, while OpenClaw actually performs operations on your machine and accounts, even autonomously.
Large language model – The language model is the thinking "brain", while OpenClaw is the body and hands built around it, capable of working with many different models.
coding agent – The coding agent specializes in programming tasks, while OpenClaw is a general personal assistant for daily administration.

Related concepts

Broader

AI-agent, Personal Assistant, Open-source software

Narrower

Skill, ClawHub

On the edge

coding agent, Prompt injection, Computer use, Model Context Protocol, Large language model, Moltbook

A brief history

Peter Steinberger, an Austrian developer, launched it in November 2025 under the name Clawdbot. At the end of January 2026, due to an Anthropic trademark objection, it became Moltbot, and a few days later, by the developer's own decision, OpenClaw. Within weeks, it became one of the most popular projects on GitHub. In February 2026, Steinberger announced that he was joining OpenAI, and the project continues as open source under a foundation.

Where things stand

Widely used among developers and technical enthusiasts, it is considered a prime example of personal agents. Due to security risks and cumbersome setup, it is not yet common among average users.

Status: 2026 November

Also known as

OpenClaw · Open Claw · Clawdbot · Moltbot · OpenClaw agent · OpenClaw AI-asszisztens · OpenClaw agent

Sources

OpenClaw – Wikipedia (angol)

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.