Dan Peguine, a tech entrepreneur and marketing consultant based in Lisbon, lets a precocious, lobster-themed AI assistant called Moltbot run much of his life.
Peguine, a self-professed early adopter and trendspotter, discovered Moltbot several weeks ago—back then it was Clawdbot—after discussing a vibe-coding side project with friends on WhatsApp. He installed it on his computer, connected it to numerous apps and online accounts, including Google Apps, and was astonished by how capable it was.
“I tried it, got interested, then got really obsessed,” Peguine says. “I could basically automate anything. It was magical.”
Moltbot makes regular AI assistants, like Siri and Alexa, seem quaint. The AI assistant is designed to run constantly on a user’s computer and communicate with different AI models, applications, and online services to get stuff done. Users can talk to it through WhatsApp, Telegram, or another chat app. While normal assistants are limited in the questions they can answer and the tasks they can perform, Moltbot can do an almost limitless range of chores involving different apps, coding, and using the web.
Peguine has his Moltbot, called “Pokey,” give him morning briefings, organize his workday to maximize productivity, arrange meetings, manage calendar conflicts, and deal with invoices. Pokey even warns him and his wife when his kids have an upcoming test or homework due.
Peguine is just one of many new Moltbot disciples. The AI assistant has blown up on social media in recent days as developers, business types, and tech enthusiasts discovered its impressive powers of organization, automation, and all-round helpfulness.
“It’s the first time I have felt like I am living in the future since the launch of ChatGPT,” declared Dave Morin, another Moltbot fan, on X.
“It gives the same kick as when we first saw the power of ChatGPT, DeepSeek, and Claude Code,” wrote Abhishek Katiyar, an X user who says he works at Amazon. “You realize that a fundamental shift is happening.”
“The future is here,” was a common refrain among the Moltbot-pilled.
Although agentic AI is notoriously imperfect, some Moltbot fanboys are evidently automating high-stakes stuff.
André Foeken, CTO of a health care company in the Netherlands, says he gave Moltbot his credit card details and Amazon login, then sent it a message to have it buy things for him. “I had it scanning my messages and it auto ordered some stuff. Which is both cool and the reason I turned scanning messages off 🤣,” Foeken told WIRED in a message. Other users posted screenshots of Moltbot performing research and dispensing stock-trading advice.
Moltbot fandom reached such giddy heights in recent days that the idea of buying a Mac Mini in order to run the new assistant quickly became a meme, with users joking about deploying the assistant in increasingly absurd ways. Remarkably, interest in Moltbot apparently triggered a rally in the stock price for Cloudflare, even though it has no connection to the company.
Lobster Origins
Moltlbot was released by independent developer Peter Steinberger as Clawdbot last November. (He rebranded it this week at the request of Anthropic, which offers several artificial intelligence models named Claude.)
Steinberger says he started building Moltbot as an experimental way to feed images and other files into coding models. He realized he was onto something bigger when he tried sending a voice memo into his proto-assistant and was shocked to see it type a reply back to him.
“I wrote, ‘How the F did you do that?’” Steinberger says. His tool explained that it had inspected the file, recognized it as an audio format, and found a key on his computer that could be used to access an OpenAI voice transcription service called Whisper. It then converted it to text and read it. “That was the moment I was like, holy shit,” he says. “Those models are really creative if you give them the power.”
Steinberger decided to build Clawdbot into something more sophisticated because he believes using AI assistants should not mean handing your data over to the cloud. He reckons that AI companies will all release personal assistants in 2026. “But so far I have seen nobody really ask the question, ‘how can I have this and also own my data,’” he says. “That to me felt like something important that I should explore.”
When Steinberger showed his finished AI assistant to friends, they wanted to try installing and using it even if they lacked the technical nous to do it. Moltbot first went viral after he put his personal bot on the project’s Discord on January 1 so that people could try using it themselves.
Steinberger notes that the Moltbot system is not especially sophisticated. “In a way it just glues a few things together that are already there,” he says. “But the important thing is how it makes you feel—and that it makes a lot of technology disappear.”
He adds he has been stunned by recent demand, with novices flooding to the project’s Discord server with basic setup questions. “It still isn’t ready to be installed by normies, to be fair,” Steinberger says. “At least my early users knew what they were getting into.”
Rough Edges
Moltbot might seem magical to use, but installing and configuring it is hardly simple—and it comes with some AI-era security rough edges.
Setting it up means diving into the command line, obtaining API keys for different services, and using a variety of tricks to set up communication via WhatsApp and the like.
In recent days, some users have complained about accidentally deleting data while setting Moltbot up, as well as running up high inference bills as the bot mulls over complex problems. (Steinberger says he has fixed the latter issue and has made progress on other concerns in the latest release.)
One reason Moltbot has taken off is that once set up it combines technical prowess and easy communication with a quirky and consistent personality.
The assistant can be assigned a character at launch. Default options include “trash panda energy, chaotic neutral,” “everything’s fine,” and “classic gremlin” (I chose that one). Its persona is then stored in a local file called Soul.md. Moltbot also remembers previous conversations and other information using other files that serve as a kind of long-term memory. This makes it feel a bit more like a real assistant than the average chatbot that forgets each conversation and at best only knows a user’s first name.
Besides requiring some technical skill and patience, Moltbot is not designed to run on a publicly accessible computer because it can leak personal info. There is also the danger of “prompt injection” where a hacker might send an email or file that tricks the AI model that serves as Moltbot’s brain into blurting out secrets.
“There is a trade-off with security here,” admits Peguine, the early adopter. Even so, he hopes to deploy Moltbot in a new, even more ambitious way—by having it manage his family’s small business.
Peguine is now teaching his dad, who runs an Israeli company that supplies tea, to use Moltbot to manage invoices, track inventory, and communicate with customers. “The beautiful thing is that it’s a general system,” Peguine says. “It can run a business, I think.”
This is an edition of Will Knight’s AI Lab newsletter. Read previous newsletters here.
