📬 This article was first published on The Nov Tech newsletter 🔗 with early access to paid subscribers. Subscribe for exclusive analysis. Medium non-members can read 2 chapters over there, including the introduction, for free.

Within 48 hours, Mac Minis began to disappear from store shelves.

Not because of a revolutionary new chip. Not because of a sale. But it was an open-source project created by an Austrian developer from his living room.

His name is Peter Steinberger. His project is ClawdBot. And it's created a massive buzz across the internet these past few days. Prepare to reconsider the full potential of AI assistants, thanks to this software.

We're not talking about an AI that answers your questions. We're talking about an agent that controls your computer, sends emails on your behalf, books your restaurants, manages your calendar, and remembers everything. Absolutely everything.

No need to open a browser or dedicated app. It functions when you message it on WhatsApp or Telegram, treating it like a colleague.

In the next few minutes, you'll understand why Silicon Valley flipped this weekend. Why some developers are declaring they've "hired their first 100% full-time employee." And most importantly, how you can do the same thing.

The Developer Who Accidentally Started a Movement

Peter Steinberger isn't unknown in the development world. He's the founder of PSPDFKit, a company that's generated millions. He was in semi-retirement when he had a simple idea: create a personal AI assistant that runs permanently on his own machine with complete system access.

The result is ClawdBot, which went from 5,000 stars on GitHub to over 20,000 in just a few days. More than 8,900 members have already joined the project's Discord, and the tech community has literally lost its mind making videos about it in every direction.

What makes ClawdBot different from ChatGPT, Claude, and other AIs you're already using is that it doesn't live in a browser tab. It lives in your messaging apps.

Like any advanced agent, you can connect it to WhatsApp, Telegram, Discord, Slack, Signal, and even iMessage. You communicate with it as you would any other contact, and it replies, but its key function is its action.

One user asked their ClawdBot to book a restaurant table. After the online booking failed, the AI employed text-to-speech to contact the restaurant and secure the reservation on its own.

Another entrusted ClawdBot with complete management of their parents' tea business: team scheduling, B2B customer tracking, inventory management, and customer service. Everything.

And that's not even the most impressive part.

ClawdBot's architecture rests on three pillars that nobody really expected from an open-source project.

The first is unlimited, persistent memory. Unlike classic chatbots that forget everything between sessions, ClawdBot stores the entirety of your exchanges locally on your machine. If you told it three weeks ago that you only drink oat milk lattes, it'll remember when you ask it to order coffee.

The second pillar is complete system access. ClawdBot can execute terminal commands, write and run scripts, install new features, and even configure other artificial intelligences. One developer recounted asking their ClawdBot to install Ollama, a tool for running AI models locally. The objective was to delegate simple tasks to a free model to save on API costs.

Result: an AI installing another AI to save you money.

The third pillar is proactivity. ClawdBot doesn't just wait for your instructions. You can configure it to send you a summary of unread emails every morning at 8 AM, alert you if a server goes down, or remind you of appointments while accounting for traffic.

The AI comes to you instead of waiting for you to come to it.

What People Are Actually Building With This

The reason for the buzz, the reason there are so many videos, what fascinates the community isn't just what ClawdBot does. It's what it can become.

Advanced users have started creating multi-agent systems. One developer configured their ClawdBot to supervise two other AIs debating each other during code reviews. And while he was walking outside, his assistant autonomously deployed a complete feature.

An even more surprising use case: a radio enthusiast gave ClawdBot access to RTL-SDR hardware, a programmable radio receiver. After just 30 minutes of zero prior training, the AI successfully analyzed the radio spectrum, identified an emergency communication system, and began deciphering real-time Fulton County firefighter conversations.

Nobody taught it how to do this.

Home automation represents another playground. Users are connecting ClawdBot to Home Assistant or even Alexa to control their homes. Health data was used by ClawdBot to automate air quality settings.

We're entering an era where your physical environment adapts to your biomarkers via local AI.

The Creator's Shocking Response to the Hype

And this is precisely why the creator's own reaction surprised everyone.

Peter Steinberger's message, released while Mac Minis were unavailable on Amazon and in European Apple Stores, went against all marketing logic. His request: that people stop buying Mac Minis.

His argument: you don't need one. A $5/month VPS does the same thing.

The official Clawd Bot account even responded to a user who was about to buy a Mac Mini, saying: "Sir, it's 2026, run Clawd Bot on a $5 VPS like a normal person."

This total transparency partly explains why the project exploded. The code is entirely open source, contributions come from everywhere, and even people who can't code can submit modifications because the essential code is itself written by AI.

The only closed part is a file called "the sauce," representing 0.001% of the project. Steinberger kept it secret intentionally as a sort of test to see if anyone would attempt to hack it.

Why This Feels Different From Every Other AI tool?

What's happening here with ClawdBot goes beyond a simple tech gadget. We're witnessing a shift.

Voice assistants like Siri or Alexa suddenly became obsolete. They respond, but ClawdBot executes. Where they forget, ClawdBot remembers. They live in closed ecosystems. ClawdBot is customizable and entirely under your control.

One analyst summarized the situation like this: "ClawdBot will destroy more startups than ChatGPT because it's self-hosted, modifiable, and belongs to nobody."

That's a significant statement, but we'll see if it's true. Traditional SaaS will struggle to compete with a free tool that performs better than most paid solutions.

We're entering a new era where everyone can have their own personal agent running 24/7, memorizing their preferences, executing their tasks, evolving with you.

The most advanced users are already talking about "vibe orchestration" rather than "vibe coding." It's the next step. They're no longer coding. They're directing a team of AIs.

How to Actually Set This Up

Setting up ClawdBot isn't as complicated as it sounds, though it does require some technical comfort. Here's the real-world walkthrough.

First, you'll need a server. Despite the Mac Mini hype, a simple AWS EC2 instance on the free tier works perfectly. You're looking at Ubuntu 24.04, and honestly, the t3. The medium free tier option with 8GB of memory is more than enough to get started.

Once you've got your AWS instance running and you've connected via SSH, the installation is literally one command. Head to the official ClawdBot site, copy the installation line (make sure you select MacOS, not Windows, in the interface), and paste it into your terminal.

The setup wizard walks you through everything. You'll choose which AI model to use as the backend. Claude from Anthropic is the best choice for demanding tasks, especially coding, even though it's pricey. For simpler stuff, you can delegate to cheaper models or even free local ones via Ollama.

Then you connect your messaging platform. I use Telegram for all my AI agents, so that's what I went with. You'll need to create a bot through BotFather (Telegram's official bot creation tool), get your token, and paste it into the setup.

Skills are where things get interesting. The Clawd Hub has a growing collection of community-built capabilities.

Want a web search? Install the Brave Search skill.

YouTube summaries? There's a skill for that.

Google Workspace integration? Available, though I'd be cautious about giving any AI full access to your Gmail just yet.

The most fascinating skill might be the self-improvement one, where ClawdBot literally learns from its own mistakes and iteratively gets better at tasks you give it.

The Real Cost Nobody Talks About

Here's what they don't tell you in the hype threads: this thing burns through tokens fast.

If you're using Claude Opus or Sonnet for everything, you can easily rack up serious API costs. That's why the smart move is to configure a hierarchy. Complex reasoning requires expensive models, while routine tasks can utilize cheaper or local options. Implement cost controls to prevent surprise bills.

The $5 VPS Steinberger recommends is real, but the AI model API costs are separate. Budget accordingly.

Also, there's the privacy elephant in the room. Yes, ClawdBot runs locally, and your data stays on your machine. But when it calls external AI APIs to process your requests, that data is going to Anthropic or OpenAI or wherever. The "everything local" pitch has an asterisk.

What This Means for the Future

I've been testing this for a few days now, and honestly? It's more complicated to manage than N8N or other automation tools I'm used to. But that's missing the point.

ClawdBot represents something bigger than workflow automation. It's the first actual glimpse of what personal AI agents will actually look like. Not assistants that live in apps. Not chatbots that forget context. But persistent, proactive, learning systems that genuinely work for you.

It's open source, that the creator is actively discouraging hardware purchases, that the community is building skills faster than any company could — this is what the AI future probably looks like.

Not dominated by big tech. Built by passionate developers who want tools that actually work.

We're still early. This interface is rough. The costs can spiral. The security implications are real. But the potential? That's what has everyone so excited.

And that's why Mac Minis keep disappearing from shelves, even though you absolutely don't need one to run this thing.

Thanks for reading. Please share your thoughts in the comments. If these advances in artificial intelligence fascinate you, and you're wondering how these technologies will transform our world in the coming years, you're in the right place. Follow me here and subscribe to my newsletter to stay updated. See u soon.

(Originally published at https://www.thenovtech.com)

References: Clawd Bot & ClawdHub, Clawd Bot GitHub reference page (creator & history), ClawdHub — community skills registry, PSPDFKit, AWS EC2, Telegram BotFather documentation, Anthropic Claude API, Ollama, RTL‑SDR, Home Assistant, n8n — workflow automation