A complete walkthrough on what OpenClaw actually is, how it became the fastest-growing repository in GitHub history, and how to get it running safely on your own hardware.
In November 2025, an Austrian developer named Peter Steinberger pushed a small project to GitHub called Clawdbot. It had a red lobster mascot, a tagline that read “The lobster way,” and a fairly simple premise: let the messaging apps you already use, WhatsApp, Signal, iMessage, Telegram, Discord, talk directly to an AI model running on your own machine, not somebody else’s cloud server.
Three days later, after a trademark dispute, it was renamed OpenClaw. That rename didn’t slow anything down. If anything, the public drama around it amplified the attention. What followed was one of the strangest growth curves in software history: from roughly 9,000 stars to over 195,000 in 66 days, a pace one contributor described as 18 times faster than Kubernetes ever grew. By mid-2026, the project had crossed 347,000 GitHub stars, making it, by star count, the most-starred repository on the platform, ahead of long-established giants like VS Code and React.
If you’ve been hearing the name everywhere and wondering what it actually does, and whether it’s worth running on your own machine, this guide walks through both.
What OpenClaw Actually Is
OpenClaw is a free, open-source personal AI agent that you run on your own devices rather than through a browser tab or someone else’s SaaS platform. Instead of waiting passively for you to open a chat window, it operates as a local-first “Gateway”, a single control plane that manages your sessions, connected channels, and available tools continuously, even when you’re not actively talking to it.
The idea Steinberger describes is “brains and muscles.” OpenClaw itself is the muscle: it’s the part installed on your machine that can read and write files, execute terminal commands, and control a web browser through the Chrome DevTools Protocol. The brain is whichever large language model you connect it to, Claude, GPT, or a local open-weight model running through Ollama, which OpenClaw calls out to for actual reasoning.
Where it gets interesting is the channel support. OpenClaw doesn’t just live in one interface, it plugs directly into the messaging apps people already use daily: WhatsApp, Telegram, Slack, Discord, Signal, Google Chat, Microsoft Teams, iMessage (via BlueBubbles), and a growing list of extension channels. In practice, this means you can message your own AI assistant the same way you’d message a friend, and it responds from your own machine rather than a company’s server farm.
Why It Grew This Fast
Explosive growth on GitHub usually has one driving cause. OpenClaw’s had several stacked on top of each other.
The first is philosophical: at a moment when most AI products are pushing users toward hosted, subscription-based cloud services, OpenClaw offered the opposite, a genuinely local-first, self-hosted assistant with full system access on your own hardware. That resonated strongly with developers who wanted an AI agent without handing every file and conversation to a third party.
The second is scope. Early community adoption data shows people using it for email triage and summarization, automated web research, code review and DevOps alerts, personal finance tracking, and daily briefings, saving reported hours per week across these categories. It wasn’t a novelty chatbot. People were plugging it into actual daily workflows within days of installing it.
The third, less flattering factor, is security attention. In early 2026, researchers disclosed CVE-2026-25253, a high-severity vulnerability (CVSS 8.8) that allowed remote code execution on exposed OpenClaw instances, with over 30,000 vulnerable instances found exposed on the public internet. Steinberger’s own framing of the project has stayed refreshingly honest throughout: he describes OpenClaw as “a free, open source hobby project that requires careful configuration to be secure,” rather than pretending it’s enterprise-hardened by default. That transparency, paired with rapid patching, seems to have built more trust than it cost.
Setting Up OpenClaw on Your Own Machine
The standard installation path follows five steps: provisioning a machine (or using your own computer directly), installing the OpenClaw core via CLI, configuring your LLM API connection, pairing a messaging app, and installing any additional skills you want.
Step 1: Run the onboarding wizard.
The recommended setup path is running openclaw onboard directly in your terminal. This CLI wizard works on macOS, Linux, and Windows via WSL2 (strongly recommended over native Windows), and walks you step by step through configuring the gateway, your workspace, channels, and skills, rather than requiring manual configuration file editing.
Step 2: Connect your LLM.
OpenClaw itself isn’t a model, it’s the orchestration layer. During setup, you’ll connect it to a “brain,” which can be a hosted frontier model like Claude or GPT, or a local open-weight model running through Ollama if you want the entire stack to stay offline.
Step 3: Pair a messaging channel.
Once your core installation is running, you pair a channel, for example, approving a Telegram connection with a command like openclaw pairing approve telegram [CODE] generated during setup. From that point forward, messages sent to that channel route through your local OpenClaw instance.
Step 4: Install skills.
OpenClaw’s functionality is extended through a plugin system, with community-built skills distributed via ClawHub, its dedicated skill directory. Rather than bloating the core project, most new capabilities, connecting to a specific API, automating a specific workflow, are added this way.
Step 5: Run a security check before exposing anything.
Given the project’s own security history, this step matters more than it might for typical software. OpenClaw includes a built-in diagnostic, openclaw doctor, specifically designed to surface risky or misconfigured direct-message policies before they become a problem. Given that CVE-2026-25253 specifically affected exposed instances, running this check, and avoiding exposing your Gateway directly to the public internet without proper configuration, is not optional if you’re serious about running this safely.
Should You Actually Run It
If you want a genuinely local, always-on AI assistant that lives inside the messaging apps you already use, and you’re comfortable with the responsibility that comes with full local system access, OpenClaw is currently the most capable open-source option built specifically for that use case. It’s free, actively maintained by a large contributor base, and explicit about its own limitations in a way a lot of software isn’t.
If you’re deploying this in any kind of shared, enterprise, or exposed environment, treat the security configuration step as the actual first step, not an afterthought. A hobby project with 347,000 stars and full shell access is a meaningfully different risk profile than a browser-based chatbot, and the project’s own creator would be the first to tell you that.
Conclusion
OpenClaw’s rise from a weekend project to the most-starred repository in GitHub history isn’t just a growth-hacking story, it reflects a genuine shift in how people want to use AI: locally, privately, and woven directly into the messaging apps they already live in, rather than locked inside another company’s browser tab. It’s powerful, actively developed, and remarkably transparent about its own limitations for an open-source hobby project of this scale. But that same power, full local system access, shell command execution, browser control, means the setup step you’re most tempted to rush through, security configuration, is the one that matters most. Run the onboarding wizard, connect your preferred LLM, pair a channel, and before you expose anything beyond your own machine, run openclaw doctor and take its warnings seriously. Do that, and you’ll have one of the most capable personal AI assistants currently available, running entirely on hardware you control.
FAQs
Is OpenClaw free to use?
Yes. OpenClaw is completely free and open-source. You do need to bring your own LLM connection, either a paid API key from a provider like Claude or GPT, or a local open-weight model through Ollama if you want to avoid API costs entirely.
Do I need a powerful computer to run OpenClaw?
Not necessarily. Since OpenClaw is the orchestration layer and the actual reasoning happens through whichever LLM you connect, a standard laptop or a small always-on machine is enough for most personal use cases. Performance depends more on your chosen LLM than on OpenClaw itself.
Is OpenClaw safe to run on my main computer?
It can be, but it requires careful configuration. OpenClaw has full local system access by design, including shell command execution and browser control, and a known high-severity vulnerability (CVE-2026-25253) previously affected exposed instances. Always run openclaw doctor after setup and avoid exposing your Gateway to the public internet without proper safeguards.
Which messaging apps does OpenClaw support?
OpenClaw connects to a wide range of channels, including WhatsApp, Telegram, Slack, Discord, Signal, Google Chat, Microsoft Teams, and iMessage (via BlueBubbles), along with several extension channels added by the community.
Who created OpenClaw, and is it still actively maintained?
OpenClaw was created by Austrian developer Peter Steinberger and is developed in the open with a large, active contributor base of nearly 2,000 people. It continues to receive frequent updates, security patches, and new community-built skills through its plugin ecosystem, ClawHub.