Late one night in November 2025, Peter Steinberger connected Claude's API to WhatsApp. The whole process took an hour. He called it a "toy".

Three months later, Meta's Zuckerberg DM'd him on WhatsApp, and OpenAI's Sam Altman used Cerebras compute resources to court him. This "toy" became the fastest-growing open source project in GitHub history—OpenClaw.

The numbers speak for themselves. Steinberger paid $10,000-$20,000 out of pocket monthly to run the project, funneling sponsorship revenue to dependency libraries, taking nothing himself. Meanwhile, OpenAI spent $13 billion of Microsoft's money. The result: this solo developer's agent framework is more popular than any product OpenAI has released.

Altman called him a "genius with amazing creativity," but there's deeper meaning behind those words. This acquisition proves a fact: the agent layer doesn't need to be built by model providers. Any developer with an API key and a messaging app can build a more compelling agent experience than big corporations. Steinberger did it.

"OpenClaw will exist as an open source project in a foundation"—that sounds reassuring, until you remember Chrome has Chromium. The open source version gets maintained, while the real agent capabilities get integrated into ChatGPT's product roadmap.


## This Isn't a Chatbot, It's an Agent Operating System

OpenClaw's core concept is clear: treat AI as an infrastructure problem, not a prompt engineering problem. It doesn't try to make the LLM "remember" context or run safely through clever prompting, but instead builds a structured execution environment around the model.

The system uses a hub-and-spoke architecture. The Gateway acts as the control plane, connecting all messaging platforms (WhatsApp, Telegram, Discord, etc.) and the AI runtime. Your messages from any platform are routed to the same persistent agent, with conversation state and tool access managed centrally on your own hardware.

One user commented under the tweet: "If anyone needs to deploy OpenClaw in the cloud, hit me up. You don't need a $10,000 Mac Studio; mine runs fine on a $25 virtual machine."

## Security Is Designed, Not Prompted

OpenClaw's security architecture is worth a close look. Each session type has different permissions and sandbox rules: the main session can run tools on the host, while private chat sessions via WhatsApp default to a stricter permission list and Docker isolation.


The Gateway binds only to 127.0.0.1 by default. Local connections are auto-approved, while remote connections require challenge-response signatures and explicit approval. Tool calls might execute in a Docker sandbox based on the session's sandbox policy—even if someone tries to exploit a prompt injection vulnerability, the blast radius is contained.

## Canvas: An Agent-Driven Visual Workspace

Canvas runs as a standalone server process on port 18793 by default. This separation from the main Gateway provides isolation (if Canvas crashes, the Gateway keeps running) and establishes a distinct security boundary.

The agent calls the Canvas update method. The server receives HTML and parses the embedded A2UI attributes, then pushes the content to the connected browser client via WebSocket. When the user clicks a button, the client sends an action event to the Canvas server, which forwards it to the agent as a tool call.

## The Irony of the Acquisition Itself

Steinberger previously sold PSPDFKit for over $100 million. He spent three years traveling and experiencing Ayahuasca, returned, failed 43 projects, and then on the 44th project built the most important open source AI agent.

OpenAI hired the person who proved you don't need $10 billion to build the future of agents. You just need to ship products faster than a committee can approve product specs.

Another reaction under the tweet was direct: "This is incredible. I've been expecting this collaboration. Now hurry up and integrate Claw into ChatGPT."


Steinberger's experience feels like a silent mockery of big tech's innovation speed. One person, at 5 AM, in one month, wrote a product more popular than companies with thousands of engineers and billions in compute resources. And now, the person who proved "small teams can disrupt big companies" has been absorbed by a big company.

One user put it more bluntly: "Signs of an acqui-hire, huge catch. Hope it stays open source forever."

Technically, OpenClaw's architecture is indeed solid. Multi-agent routing allows directing different channels or groups to completely isolated agent instances, each with its own workspace, model, and behavior. This means you can set up a helpful moderator persona for a Discord server bot, while configuring different tool access and a more formal tone for Telegram support DMs.

But the most intriguing part is that the story's ending isn't an open source project defeating a big company, but a big company neutralizing the threat through acquisition. Steinberger proved the speed advantage of small teams, and then OpenAI solved the problem in the way he knows best—writing code—by acquiring the person who writes the code.