🦞AIClawGuide
Community Guide

What Is OpenClaw?

A plain-language introduction to OpenClaw as a self-hosted AI agent, including what it does, why it exploded, and where the official GitHub repo lives.

Mar 13, 2026

The short version

OpenClaw is an open-source, self-hosted AI agent system. The current official README frames it as a personal AI assistant you run on your own devices, with the Gateway as the control plane rather than the whole product.

The Orange Paper frames the project in one sentence that is still useful:

  • ChatGPT is a consultant
  • OpenClaw is an operator

That difference matters because OpenClaw is designed to stay online, connect to real channels, call tools, keep memory, and work through a gateway you control.

Is OpenClaw an AI agent?

Yes. If you are searching for openclaw agent or openclaw ai agent, the simplest accurate answer is:

OpenClaw is a self-hosted AI agent platform built around an always-on Gateway, persistent workspace state, tool use, memory, channels, and device nodes.

That is different from a normal chat app in a few ways:

  • it is meant to keep running
  • it can route through real message surfaces
  • it can call tools and nodes
  • it can keep durable state in workspace files
  • it expects the operator to control runtime, auth, and risk boundaries

So “AI agent” is the right category, but it is more useful to think of OpenClaw as an operational agent system rather than a single prompt window.

Where is the official OpenClaw GitHub repo?

If your real intent is openclaw github or github openclaw, the official repository is here:

The two most useful official links to keep together are:

This page is on purpose not the official repo. AIClawGuide is an unofficial community guide layer that helps people understand and navigate the official project faster.

Core snapshot as of March 11, 2026

The current Orange Paper snapshot highlights a project that moved unusually fast:

  • 280,000+ GitHub stars
  • 53,232+ forks
  • 1,075+ contributors
  • 13,729 registered ClawHub skills
  • 55 bundled skills
  • 20+ supported message channels
  • recommended stable line: v2026.3.8

Those numbers are part of why OpenClaw no longer fits the “interesting side project” category.

How it differs from a hosted chat product

Hosted chat tools answer questions inside a product someone else runs. OpenClaw changes the control surface:

  • you choose the model
  • you choose the runtime
  • you choose the channels
  • you choose the skills
  • you own the data and operational risk

That last point is important. Self-hosting gives you control, but it also makes security and cost control your responsibility.

Why people install it

Most people are not installing OpenClaw because they want “more AI.” They install it because they want a different kind of AI:

  • a system that can live in Telegram, QQ, Feishu, Discord, or WhatsApp
  • a system that can keep state and memory across sessions
  • a system that can switch between Claude, GPT, Gemini, DeepSeek, GLM, Kimi, MiniMax, and local Ollama
  • a system that can grow by adding skills

In other words, the appeal is not only model quality. It is routing, extensibility, persistence, and control.

Why the community calls it “raising lobsters”

OpenClaw inherited a lobster mascot from its earlier names. In Chinese communities, running and maintaining an OpenClaw instance quickly became known as “raising lobsters.” The phrase spread because it makes a technical project feel social and legible.

That culture is part of the project’s growth story, but it also hides a serious operational truth: raising a lobster means maintaining a live system with real costs and real security boundaries.

What AIClawGuide is

AIClawGuide is not the official OpenClaw website.

It is an unofficial community guide that reorganizes the source material into operator-friendly paths:

  • deployment
  • channels
  • models
  • skills
  • security and cost
  • comparisons and ecosystem context

When we reference official documentation or official sponsor information, we say so directly. Everything else on this site is community-built editorial guidance.

The right way to start

The Orange Paper strongly implies a sequence that removes confusion:

  1. Get a local install working.
  2. Configure one model path only.
  3. Add one easy channel, usually Telegram.
  4. Learn security and budget basics before you expand.
  5. Add skills only after the base loop is stable.

That order is much safer than jumping straight into a public cloud deployment with multiple channels and random third-party skills.

Where to go next