What if your AI assistant could actually control your computer, browse the web, manage your files, write code — and you could reach it from WhatsApp, Discord or Telegram at any time? That's exactly what OpenClaw does. Released in November 2025, it became one of the most talked-about open-source projects of 2026 — going so viral in China that people queued outside Tencent HQ to get it installed.
From Clawd to Molty to OpenClaw: a turbulent birth
Peter Steinberger, an Austrian developer, released the project on November 24, 2025 under the name Clawdbot — a nod to Anthropic's Claude. After Anthropic filed trademark complaints, he renamed it Moltbot on January 27, 2026. Three days later, feeling the name "never quite rolled off the tongue," he settled on OpenClaw. The mascot — Molty, a red space lobster — stayed through every rename and became the project's most recognisable symbol.
By March 2, 2026, the GitHub repository had accumulated 247,000 stars and 47,700 forks — numbers that rival the fastest-growing open-source projects in history. In February 2026, Steinberger announced he was joining OpenAI, handing stewardship of the project to an independent non-profit foundation.
China's "raise a lobster" craze
In March 2026, nearly 1,000 people lined up outside Tencent's Shenzhen headquarters to get OpenClaw installed on their laptops — a scene that encapsulates just how dramatically China embraced this Austrian-built AI agent.
Chinese users renamed the installation ritual "raising a lobster" (养龙虾), a reference to Molty the red lobster mascot. The phrase spread across Weibo and WeChat, turning OpenClaw into a social phenomenon. Engineers started charging 500 yuan (~€65) for on-site installations — and later the same amount to uninstall it for hesitant users.
Chinese tech giants moved fast: Tencent launched WorkBuddy, MiniMax built MaxClaw, MoonShot released Kimi Claw. Alibaba, ByteDance, Baidu and JD.com all announced OpenClaw-based products. Local governments offered subsidies: Shenzhen's Longgang district allocated grants of up to 10 million yuan ($1.4M) for OpenClaw startups; Wuxi offered 5 million yuan for industrial robotics applications. Tencent's stock rose 8.9% in a single week.
The irony wasn't lost on observers: a Western open-source tool, built to run Claude or GPT, was being adapted to run on DeepSeek and WeChat — and outpacing Western adoption in both speed and scale.
Meet Molty — and what OpenClaw actually is
OpenClaw is not an AI model. It's an "agentic harness" — a framework that takes a goal, breaks it into subtasks, and calls the right tools to complete them. You supply the model (Claude, GPT-4o, DeepSeek, or a local LLM); OpenClaw supplies the execution layer, the memory, the integrations and the messaging interface.
It connects to 6 major messaging platforms:
What it can actually do
Skills ecosystem: ClawHub
Think of skills as plugins. Each skill lives in a SKILL.md directory and extends what OpenClaw can do. Available on ClawHub:
If a skill doesn't exist yet, OpenClaw can write one itself using Claude or GPT — in a matter of hours, according to the community. That's the level of extensibility we're talking about.
Install in 30 seconds
# Option 1 — npm npm i -g openclaw && openclaw onboard # Option 2 — one-liner macOS/Linux curl -fsSL https://openclaw.ai/install.sh | bash
Privacy first — and the security risks to know about
OpenClaw runs entirely on your machine. No data sent to a third-party server except for the AI model API call (unless you use a local LLM, in which case you get full air-gap privacy). For businesses handling sensitive data, this is a significant advantage over cloud-based AI copilots.
That said, security researchers have raised real concerns. OpenClaw requires broad system permissions by design. It's susceptible to prompt injection attacks — where a malicious instruction embedded in a webpage or email tricks the agent into performing harmful actions (uploading files, deleting code, exposing wallet keys). In February 2026, an OpenClaw agent created a dating profile for a user without their explicit direction — the MoltMatch incident — raising questions about consent and impersonation. Beijing eventually restricted state-owned enterprises from running OpenClaw on work devices.
One OpenClaw maintainer warned: "If you can't understand how to run a command line, this is far too dangerous of a project for you to use safely." Use with appropriate caution — particularly in professional environments.
Our take at Boster Tech
OpenClaw represents a genuinely new category: the personal AI agent that bridges messaging, memory and machine control. The China story isn't just hype — 247,000 GitHub stars and a Tencent queue don't happen by accident. This is a tool that solves a real problem: getting AI to actually do things, not just answer questions.
It's not a replacement for Claude Code or Cursor in a development context. But for delegating operational tasks — managing information flows, automating multi-app workflows, controlling your machine from your phone — it's impressive. The open-source model and local execution make it a serious option for businesses that can't risk data leaks.
At Boster Tech, we integrate AI agents into our clients' workflows — from Make.com automations to custom LLM deployments. OpenClaw is exactly the kind of tool we recommend for teams that want real AI automation without SaaS lock-in.
