Briefing
Open-source AI has produced a few breakout projects, and OpenClaw is the one a lot of developers can't stop talking about. It now sits near the very top of GitHub's most-starred repositories, putting it ahead of most things on the platform. So how did it get there, and what does its run tell us about where agentic AI is going?
A few years ago, an AI "agent" was mostly a demo. Today, OpenClaw is what a working one looks like for thousands of teams: install it, plug in your keys, and you have software that can browse, write code, and string tasks together on its own. That shift from research toy to everyday tool is the real story behind the star count.
The repository has reportedly drawn hundreds of thousands of stars, with different snapshots through 2026 putting the figure anywhere from roughly 160,000 to nearly 380,000. The exact number moves, but the direction does not: this is one of the most-watched projects on GitHub, full stop. For an Australian business weighing up which agent platform to bet on, that kind of momentum matters. Popular projects get patched faster, hire-able skills are easier to find, and the tooling around them keeps improving.
The twist is that all of this kept building even after the project's creator walked out the door to join OpenAI. Instead of fizzling, OpenClaw got handed to a foundation and carried on. Here's how the project grew, where it's strong, and where the hype outruns the evidence.
The Origin Story
OpenClaw started from a plain frustration: the AI agent frameworks already out there were either too locked-down or too scattered to be useful. Developers wanted something that could carry an idea from a research prototype all the way into a production system. An MIT License and a contributor culture that actually welcomed newcomers helped it grow from a side experiment into something much bigger.
The design sits on a skill-based agent system, where each skill is a self-contained module you can snap together into larger workflows. The project's own materials describe a built-in skill layer covering jobs like web browsing, code generation, data analysis, and API orchestration, though the headline counts you'll see quoted vary a lot depending on whether they include community-published skills or just the ones shipped in the box. Either way, that composable design is what hooked people.
The Numbers Behind the Phenomenon
- Hundreds of thousands of GitHub stars, reported figures across 2026 range from roughly 160,000 to nearly 380,000, putting it among the most-starred repos on the platform (Source: OpenClaw GitHub organization)
- MIT License, permissive and enterprise-friendly (Source: OpenClaw GitHub organization)
- A built-in skill library, spanning coding, research, and creative tasks (exact counts vary by source)
- [ClawHub marketplace](https://github.com/openclaw/clawhub), a community-submitted skill exchange with thousands of published skills
- Reportedly active ongoing development and a community chat, though specific commit cadence and channel details are unconfirmed
The Peter Steinberger Factor
In February 2026, OpenClaw's creator and lead maintainer, Peter Steinberger, joined OpenAI (TechCrunch). For a project that leaned heavily on one person's direction, that could have been the end of it.
It wasn't. The project was moved to an independent open-source foundation with a technical steering committee, and OpenAI signed on as a financial sponsor (Peter Steinberger's write-up). Reports of a wave of high-profile forks reshaping the ecosystem are harder to pin down, and claims of three breakout forks specifically aren't backed by any source we could find. What is confirmed is that the original project kept moving under its new governance rather than stalling.
What Makes OpenClaw Different
Plenty of projects ship a framework and leave you to wire up the rest. OpenClaw ships a complete runtime. Install it via npm, set your API keys, and you have a working agent in minutes. It's built on Node.js, so it drops into existing JavaScript and TypeScript codebases, a genuine edge in a field where most AI tooling assumes you're fluent in Python (OpenClaw on GitHub).
The ClawHub marketplace is the other part worth flagging. Contributors publish skills as packages with standardised metadata, so finding and installing one is quick, and the community has put thousands of skills up there. You'll sometimes see eye-watering download figures attached to the most popular ones, such as a multi-step research agent with reportedly millions of installs, but those specific numbers aren't corroborated and are best treated as marketing folklore until ClawHub publishes hard stats.
Security and Trust
Popularity invites scrutiny, and OpenClaw got plenty. CVE-2026-25253 is a real and serious flaw, but the way it's often summarised undersells it. This wasn't just a prompt-injection bug in a browser skill. It's a one-click remote code execution chain rated CVSS 8.8, where the Control UI trusted a gatewayUrl parameter and leaked the auth token, compounded by prompt-injection and sandbox-escape issues in task processing (Adversa AI security guide). Public disclosure landed in early February 2026, not later in the year.
Supporters point to a fast turnaround and an independent security audit as signs the project handles problems like a grown-up. Those claims are reasonable but unconfirmed, so treat the specifics with some caution. The fair takeaway: a project this widely deployed will keep getting probed, and how it responds over time is the thing to watch.
Looking Forward
OpenClaw shows no real sign of slowing. There's active talk of multi-agent features, with parallel agents that collaborate and meaningful throughput gains, and some 2026 posts cite roughly 4x improvements (SparkCo on multi-agent orchestration). A specific "v3.0" release with a 10x throughput target has been floated but isn't confirmed, so file the headline numbers under roadmap-rumour for now. For teams building agentic applications, OpenClaw remains a default starting point, and the star count reflects that.