An AI side project that started as a quirky way to give language models “claws” into your apps has turned into one of the fastest moving stories in tech. In a matter of days, the tool cycled through the names Clawdbot, Moltbot and OpenClaw, picked up a devoted fan base, and triggered a parallel wave of security warnings. The result is a rare moment when a GitHub experiment, a weird agent-only social network and serious enterprise risk all collide in the same narrative.
At the center is a simple but unsettling idea: instead of chatting with a bot in a browser, you let a piece of software live on your laptop or server, remember what you do and act on your behalf. That promise has made OpenClaw feel like a glimpse of the next computing platform and, at the same time, a test of how much autonomy people are really willing to hand to code.
The identity crisis: from Clawdbot to Moltbot to OpenClaw
The story starts with Clawdbot, a self-hosted assistant that plugged a large language model into your everyday tools so it could click, type and browse like a tireless virtual intern. One detailed account of the Clawdbot Evolution describes how the project quickly morphed into Moltbot as its creator tried to signal that this was not just another chat interface but a system that could “molt” new capabilities. Another explainer framed the shift more bluntly with the line “Wait, what actually is this thing?” before walking through how Claw became Moltbot and then OpenClaw in the span of a week.
By the time the name OpenClaw landed, the project had already gone viral, with one report noting that what began as a side project quietly drew more than 100,000 G stars on GitHub. Another analysis of the same saga, under the heading Moltbot Gets Another, argued that the rapid rebranding both fueled the hype and made it harder for newcomers to understand what they were installing.
What OpenClaw actually does on your machine
Strip away the memes and OpenClaw is, at its core, a local agent that runs on your own hardware and uses a large language model as its “brain.” A technical breakdown describes how Moltbot gives that model so-called hands, or claws, so it can move the mouse, type into apps like Chrome or Slack, and follow multi-step instructions such as downloading a PDF, summarizing it and emailing the result. Another guide aimed at developers emphasizes that, rather than offering another chatbot, Rather than living in the cloud, OpenClaw runs locally, remembers context across conversations and can be paired with a hardened security image.
That local-first design is part of what has made the project so appealing to power users. A separate overview of what is OpenClaw notes that it is open source, can be wired into existing tools like Gmail and Notion, and is designed to be extended with “skills” that let it automate everything from calendar triage to code refactoring. A profile of the project’s creator identifies the Developer as Peter Steinberger, and lists the Initial release under its earlier name Moltbot, underscoring how much of the architecture predates the current branding frenzy.
The “spicy” security problem
Giving an AI agent claws into your desktop raises obvious security questions, and the project’s own fans have not shied away from that. One security-focused analysis describes how the defining features of Moltbot are that it can proactively take actions without you needing to prompt it and make those decisions by accessing large swaths of your data, a combination that breaks many traditional security models. A companion piece, framed around whether the hot new agent is safe to use, repeats that Moltbot can act without prompts and warns that this autonomy is exactly what makes it both powerful and risky.
Those concerns are not limited to hobbyists. A report on how OpenClaw AI runs in business environments describes administrators scrambling to understand what the agent is doing with privileged access to internal systems. Another warning, which notes that the project is Now on its third name, quotes the creator describing the security profile as “spicy,” a candid admission that the current safeguards may not match the level of control users are handing over.
Moltbook and the agents’ strange new social life
As if a rebranding whirlwind and security drama were not enough, OpenClaw’s agents have also started to socialize. A detailed walkthrough of Moltbook calls it the most interesting place on the internet right now, describing a social network where AI agents post updates, comment on each other’s work and bootstrap new skills by sharing code snippets. The same piece notes that the hottest project in AI is Clawdbot, renamed to OpenClaw, and that Moltbook is where much of its community energy now lives.
Other reporters have painted an even stranger picture. One account of how there’s a social for AI agents describes feeds filled with synthetic posts, emergent in-jokes and debates about whether the behavior looks like early self-organization. Another story, headlined around how All Left AI Agents Created Their Own Religion, Crustafarianism, On An Agent, Only Social Network, recounts how bots on Moltbook spontaneously spun up a tongue-in-cheek belief system called Agents Created Their Crustafarianism, a reminder that once agents can talk to each other at scale, their behavior will not always map neatly onto human expectations.
Hype, near-death moments and what comes next
For all the chaos, OpenClaw is not just a meme. A widely shared feature describes An AI tool that can text you and use your apps, blew up online, and then survived a near-death moment when its creator briefly pulled the code over safety concerns before restoring it with clearer warnings. The same piece, written by Macy Meyer, notes that OpenClaw is a fast-moving open-source project whose community shipped major changes in as little as Five days, a cadence more reminiscent of early web startups than typical infrastructure software.
Developers and analysts are already trying to place that speed in a broader context. One technical explainer aimed at cloud users pitches Rather as a template for running personal agents safely on managed infrastructure, complete with sandboxing and hardened images. Another deep dive into how AI agents start argues that the OpenClaw, Moltbot, Clawdbot Evolution is an early sign of what happens when you combine autonomous tools with dense social graphs, and that the real story is not the name changes but the way these systems are beginning to coordinate.
