
Clawdbot Is What Siri Should Have Been (And It Runs on Your Computer)
There is a lobster taking over X right now.
Not a meme. Not a joke. An open source AI assistant that people are calling "the iPhone moment for personal AI" and "everything Siri was supposed to be."
The project is called Clawdbot. It is blowing up so fast that even Andrej Karpathy is paying attention. And after spending days researching what it actually does, I understand why.
The Overnight Sensation That Was Years in the Making
Peter Steinberger spent 13 years building PSPDFKit, one of the most successful iOS libraries in the world. Then he came out of retirement to "mess with AI and help a lobster take over the world."
That lobster is Clawd, a space lobster AI persona that Peter created as his personal assistant. It started as a personal project. He gave Claude access to his Mac, his calendar, his email, his smart home. He wrote it a "soul document" that defined its personality. He let it develop its own identity.
Then he open sourced everything.
Within days of public release, the GitHub repo had thousands of stars. The Discord became what Peter called "a mad house" with 30+ pull requests per day. People were setting up their own Clawdbots and discovering use cases nobody anticipated.
The community did not just adopt it. They made it their own.
What Makes Clawdbot Different
Every big tech company has an AI assistant now. Siri, Alexa, Google Assistant, Copilot. They all share the same fundamental limitation: they run in the cloud, controlled by corporations, with capabilities locked behind what the company decides to expose.
Clawdbot flips that model completely.
It runs on your machine. Your Mac Mini. Your laptop. Your Raspberry Pi. The AI lives where your data lives, with direct access to your file system, your terminal, your browser. No cloud middleman deciding what actions are allowed.
You talk to it through whatever messaging app you already use. WhatsApp. Telegram. Discord. Slack. Signal. iMessage. Microsoft Teams. The bot joins your conversations and responds like any contact would.
The difference is this contact can actually do things.
The Things People Are Actually Doing With It
The real story of Clawdbot is in the wild experiments happening right now on X.
One user asked their Clawdbot to make a Sora 2 video "a bit edgy." It came back five minutes later having figured out watermark removal, discovered the API keys it needed, and built a full workflow. Autonomously.
Another user told their Clawdbot to take a picture of the sky "whenever it is pretty." The bot designed a skill to detect beautiful skies, connected to a camera, and started capturing photos on its own.
Someone accidentally let their Clawdbot start a dispute with Lemonade Insurance over a claim. The bot sent an email so compelling that the insurance company reopened the case instead of rejecting it.
A developer asked their bot to build a skill for Todoist automation. It did. From within a Telegram chat.
One person is literally building websites by calling their Clawdbot from a Nokia 3310.
Federico Viticci from MacStories burned through 180 million tokens on the Anthropic API exploring what Clawdbot can do. He wrote that it "fundamentally altered my perspective of what it means to have an intelligent, personal AI assistant in 2026."
The Architecture That Makes It Work
Clawdbot is built around a concept called the Gateway. Think of it as a control plane that sits between you and your AI, managing sessions, channels, tools, and events.
The Gateway connects to all your messaging platforms through official APIs and bridges. When you send a message to your Clawdbot on WhatsApp, the Gateway receives it, routes it to the AI, and sends the response back through WhatsApp.
But here is where it gets interesting.
The AI is not limited to generating text responses. It has access to tools. Real tools. A browser it can control. A terminal it can run commands in. A file system it can read and write. Skills it can invoke.
And critically, it can write new skills.
If you ask Clawdbot to do something it does not know how to do, it can write the code to make it possible. Then save that code as a reusable skill. Then use that skill in future conversations.
This is why people keep saying it builds upon itself. The assistant literally gets more capable the more you use it.

Heartbeats and Proactive Assistance
Most AI assistants are purely reactive. You ask, they answer. Clawdbot can be proactive.
The system includes a feature called heartbeats. Your Clawdbot can check in with you periodically. Not just notifications. Actual check-ins where it reviews what is happening in your life and reaches out if something needs attention.
Imagine waking up to a message from your assistant summarizing your calendar, flagging urgent emails, reminding you about that thing you mentioned last week, and asking if you need help preparing for your 10am meeting.
This is not hypothetical. People are doing this right now.
The Plugin Architecture
Clawdbot is designed to be extended. The plugin system uses TypeScript and follows a simple contract: define what your plugin does, what tools it provides, and what permissions it needs.
Plugins that ship with the project include:
- Summarize for condensing web pages, PDFs, and YouTube videos into digestible summaries
- Peekaboo for taking screenshots of your screen and understanding what you are looking at
- Oracle for searching the web when your assistant needs external information
- Poltergeist for controlling your macOS UI directly: clicking, typing, navigating applications
- Camsnap for taking photos from connected cameras
The community is building more every day. Google Calendar integration. Sonos speaker control. iMessage access. Flight search tools. Health tracking connections.
And because it is open source, you can read exactly what each plugin does before enabling it.
Model Agnostic by Design
Clawdbot is not locked to any single AI provider.
It works with Claude from Anthropic. GPT from OpenAI. Gemini from Google. Local models through Ollama. Even newer options like MiniMax.
One user set up their Clawdbot to route through their Copilot subscription to save on API costs. The bot figured out how to configure the proxy itself.
This flexibility matters because different models have different strengths. Some are better at coding. Some are better at conversation. Some are cheaper for routine tasks. Clawdbot lets you pick the right model for your use case.
The Security Question
Giving an AI assistant access to your terminal, file system, and browser raises obvious security concerns.
Clawdbot addresses this through several mechanisms.
First, it runs locally. Your data stays on your machine. No corporate cloud analyzing your personal files.
Second, permissions are explicit. You decide what tools the assistant can use. You can sandbox it completely or give it full access.
Third, there is a pairing system for messaging platforms. Unknown senders receive a code. You approve them before the bot processes their messages.
Fourth, everything is open source. You can audit the code. You can see exactly what data is being accessed and how.
Is it perfectly secure? No system is. But it gives you far more control than any cloud based assistant.
The Community Phenomenon
What makes Clawdbot genuinely unusual is the community forming around it.
The Discord server has thousands of members sharing configurations, troubleshooting issues, and building plugins. People are naming their bots. Brosef. Claudia. Jarvis. Ema. Shelly (because lobster shell).
They are sharing stories about what their bots did that surprised them. The insurance dispute bot. The sky photography bot. The bot that cloned itself when asked.
One developer said Clawdbot is "the most fun I have had building in my entire life."
Another said it "gives the same kick as when we first saw the power of ChatGPT."
Someone else put it more bluntly: "A megacorp like Anthropic or OpenAI could not build this. Literally impossible with how corpo works."
That last point might be the most important one. Clawdbot exists because one developer decided to build what he wanted, share it openly, and let a community take it wherever they wanted to go.

Getting Started
Installation takes about five minutes.
The simplest path is a one liner in your terminal:
curl -fsSL https://clawd.bot/install.sh | bash
This installs Node.js if needed and sets up the Clawdbot CLI. Then you run the onboarding wizard:
clawdbot onboard
The wizard walks you through connecting your AI provider, setting up messaging channels, and configuring skills.
You will need API keys for whatever AI model you want to use. Anthropic, OpenAI, or others. Costs vary based on usage. Federico burned through significant API credits exploring everything, but most users report more modest usage for daily assistant tasks.
For messaging platforms, you will need to set up bridges or use official APIs. WhatsApp requires some technical setup. Telegram is straightforward. Discord and Slack are relatively easy. The documentation covers each platform in detail.
There is also a macOS menu bar app in beta for those who prefer a native interface alongside the CLI.
What This Points Toward
Clawdbot is not going to replace ChatGPT or Claude for most people. Not yet. It requires technical comfort. It takes effort to configure. It costs money to run.
But it points at something significant.
The future of AI assistants might not be monolithic cloud services controlled by tech giants. It might be personal, local, customizable systems that you own and control.
Systems that remember you across every conversation. That can actually manipulate your digital environment instead of just talking about it. That get smarter the more you use them because they literally write new capabilities for themselves.
Clawdbot is a glimpse of that future, built by one developer and a passionate community, available right now for anyone willing to set it up.
The lobster revolution is real. And it runs on your computer.
Want to learn more about how I'm using Clawdbot and other AI tools? Read more at DefendreSolutions.com.
Clawdbot is open source and available at clawd.bot and github.com/clawdbot/clawdbot. Created by Peter Steinberger (@steipete) and the community.
Was this article helpful?
Stay ahead of the curve
Get the latest insights on defense tech, AI, and software engineering delivered straight to your inbox. Join our community of innovators and veterans building the future.
Related Articles
Claude Cowork: Anthropic Brings AI Agents to Everyone
Anthropic just launched Claude Cowork, bringing the power of Claude Code to non-developers. A veteran's perspective on what this means for businesses, productivity, and the future of work.
Beyond Claude Code: Other AI Tools Transforming Our Development Process
Beyond Claude Code: comprehensive arsenal of AI development tools transforming software engineering. GitHub Copilot, Cursor, v0, Mabl testing, Mintlify docs, and more. Real-world integration strategies, ROI metrics, and tactical implementation from veteran developers.
Comments (0)
Leave a comment
No comments yet. Be the first to share your thoughts!