Everything about OpenClaw (formerly Clawdbot, formerly Moltbot)
Published on
If you blinked in the last 2 weeks, you might have missed the birth, death and rebirth of the most viral AI project in recent times, ClawdBot (which has already been renamed MoltBot and is currently “OpenClaw”).
The problem is that the name “Clawd” sounded a little too much like “Claude”, and Anthropic’s legal team did not like that one bit. The project changed to MoltBot, a reference to a lobster, the product’s logo, molting its shell, but thanks to a slip-up by the creator, someone grabbed the name first, and he ended up settling on “OpenClaw”.
The speed at which this project grew is insane. As I write this post, the project already has almost 160 thousand stars on GitHub. That’s a record comparable to the launch of ChatGPT itself or DeepSeek.
And why so much hype? Because it’s the first tool with real potential for something everyone has been asking for over the past months: autonomous agents that actually do things. It’s a bot that can run straight from your local machine, 24/7, you can integrate it with everything (Gmail, ElevenLabs, Claude Code, etc.), connect it via WhatsApp (I don’t recommend it) or Telegram, and it can access your data, documents, use the browser and more. It doesn’t give you an answer, it goes and executes. If it doesn’t know how? You ask it to figure it out.

And you might (rightly) wonder: “But weren’t there already a bunch of systems that did these things?”. There were, but not at this level, and not with this much “freedom”. ChatGPT’s agent mode or Anthropic’s Computer Use have existed for a while, but while they ask you “Are you sure?” three times before executing an action, OpenClaw just goes ahead and does it.
Maybe the coolest part, which I took from a conversation with Sérgio Lopes (@sergio_caelum), is that OpenClaw doesn’t try to solve your problem by “chatting”, it translates your human problem into a computable problem. Basically, for almost anything you ask, its answer is to generate and run code (usually Python or Bash) in a loop, without being afraid to poke around and keep going (which can be both an advantage and a cause for concern).
On top of that, as it gets to know you better, it builds a Markdown index of your entire conversation history, how it solved some problem before (and it can use that knowledge to solve another one), and reaches a level of personalization that evolves to the point where, some time from now, my OpenClaw will probably be very different from yours. It’s the promise of a personalized agentic AI with no brakes… kind of a slightly reckless “vibe coding”, for better and for worse.
With that, the biggest concern is security. A lot of people are already calling it “Shadow AI”, because if you don’t lock it down properly, it will certainly increase (by a lot) the surface for potential prompt injection attacks and vulnerabilities, from your machine to your email and the products you use. That’s why the recommendation is to install it on a VPS or a dedicated machine, not your main one. And that’s also why I don’t suggest installing it on the PC of that uncle of yours who saw the hype on the news and now wants a Jarvis. There’s a big risk he’ll expose API keys and personal data without even noticing.
I installed it here on a VPS and I can tell you the initial setup process is SOOO annoying. For non-technical people (and even for technical ones), it’s pretty exhausting.
Is it dangerous to run this on your main machine with root access? VERY. But we’re seeing a first taste of a concept that will certainly become BIG products in the coming months.