How OpenClaw Became GitHub's Most-Starred Project in 60 Days

In November 2025, a developer named Peter Steinberger built a weekend project while in Madrid. Four months later, it had more GitHub stars than React, a framework with a 10-year head start.

That project is OpenClaw. And the data behind its rise is unlike anything the open-source world has seen before.

By March 2026, OpenClaw had crossed 335,000 GitHub stars, surpassing React's all-time cumulative record. It did so in roughly 60 days after its public relaunch. React took a full decade to reach 250,000.

This is not just a growth story. It is a signal. The developer community is telling us something important about where AI is heading, and how fast it is moving. This post breaks down the numbers, compares the top open-source agentic AI projects side by side, and draws out what the whole trend means for teams deciding how to build with AI in 2026.

What Is OpenClaw?

OpenClaw is an autonomous AI agent that runs locally on your devices. Unlike cloud-based AI assistants, it operates as a persistent background process with direct access to your files, shell, browser, and APIs.

What makes it genuinely different from every framework that came before it:

• It is always on. OpenClaw checks for work every 30 minutes without being prompted. You do not need to open a chat window; it is already running.
• It unifies your inboxes. It connects natively to over 15 messaging platforms including WhatsApp, Telegram, Slack, Discord, Signal, and iMessage. You instruct it from any channel; it works across all of them simultaneously.
• Your data never leaves your machine. All storage is local, in plain Markdown files. No cloud sync, no vendor lock-in, no data sharing agreement to sign.
• It supports multi-agent architectures. Multiple specialized agents can run in a single Gateway process, each with isolated memory, credentials, and tool permissions.
• It has a marketplace. ClawHub hosts over 1,100 community-contributed AgentSkills that extend what your agent can do, from booking calendars and sending emails to running code pipelines and monitoring dashboards.

The project started life as "Clawdbot" in November 2025, briefly became "Moltbot" after a trademark dispute, and settled on "OpenClaw" three days later. The rebrand drama itself generated a wave of press coverage that accelerated growth long before the product went viral on its own merits.

The Raw Numbers

To understand why those numbers matter: React is the most widely used JavaScript UI library in the world, maintained by Meta with thousands of contributors and tens of millions of developers using it daily. It took a full decade to accumulate 250,000 stars. OpenClaw hit that number in a single calendar quarter, built by one person who started coding it as a side project on a weekend.

OpenClaw vs. the Field: GitHub Stars Compared

Let us look at where OpenClaw sits relative to the five most actively discussed open-source agentic AI projects right now.

The gap is not incremental. OpenClaw has 5 times the stars of the second most popular project in the category. That kind of lead does not happen by accident or by marketing spend. It happens when a product solves a problem that the community had been waiting for someone to solve.

Growth Velocity: Stars Per Day

Total star counts can be misleading because older projects had more time to accumulate them. A fairer comparison is the rate of growth: how many new stars is each project earning per day on average?

OpenClaw is growing at 15 times the daily rate of the second-placed project. At this pace, it is not catching up to the competition; it has lapped them entirely and is on a different trajectory.

The 60-Day Rocket: A Timeline

The growth of OpenClaw is best understood as a sequence of compounding events, not a single viral moment. Each milestone generated press coverage that fed the next wave of signups.

Each milestone fed the next. The rebrand drama generated press coverage even before most developers had used the product. The Hacker News post drove the first wave of power users. Peter Steinberger joining OpenAI generated mainstream tech coverage from CNBC, Fortune, and NBC News. Jensen Huang calling it "the next ChatGPT" at a public event reached an entirely different audience: enterprise decision-makers, not just developers.

Why OpenClaw Won the Internet

Five structural factors combined to produce this outcome. Understanding them matters because they tell us what the developer community was genuinely waiting for, not just what went viral this quarter.

1. The timing was perfect

The ChatGPT wave of 2023 created massive demand for AI that could actually do things, not just chat. By late 2025, developers had spent two years building LLM-powered apps and hitting the same wall: stateless, single-session interactions that could not maintain context or take persistent action. OpenClaw solved exactly that problem at exactly the right moment.

2. Creator credibility lowered the trust barrier

Peter Steinberger founded PSPDFKit, a PDF SDK company acquired for over $100 million. When he shipped OpenClaw, his track record meant the project was taken seriously immediately. First-time contributors did not need to wonder whether the creator would abandon it after losing interest. That reduced the psychological cost of starring, forking, and building on top of the project.

3. Local-first was a genuine moat

Every other agentic AI framework requires a cloud connection, either to the LLM provider or to the framework's own infrastructure. OpenClaw runs entirely on your hardware. Your data never leaves. In an environment where enterprise security teams are blocking AI tools by default, "local-first" is not just a feature; it is a category differentiator.

4. Messaging-first UX solved a real problem

Nobody wants another dashboard. The core insight at the heart of OpenClaw is that you already live in your messaging apps. An AI agent that meets you where you are, in WhatsApp or Slack or Discord, and operates across all of them simultaneously, is fundamentally more useful than one that requires you to open a new interface. That UX insight resonated with both technical and non-technical users.

5. The controversy was free marketing

Every time a security researcher published findings about exposed OpenClaw instances or malicious ClawHub skills, it generated another wave of press coverage. Hundreds of thousands of developers who had never heard of it searched for it to understand the risk. A significant fraction stayed because they liked what they found.

Feature Comparison: OpenClaw vs. the Top 5

The stars tell one story. The features tell another.

Feature	OpenClaw	AutoGen	CrewAI	LlamaIndex	OpenHands
Always-on persistent agent	Yes	No	No	No	No
Local-first, no cloud required	Yes	No	No	No	No
Native messaging channels (15+)	Yes	No	No	No	No
Multi-agent in one process	Yes	Yes	Yes	Partial	No
Proactive scheduling	Yes	No	No	No	No
Skill/plugin marketplace	Yes (1,100+)	No	No	No	No
Production-ready enterprise docs	Improving	Yes	Yes	Yes	Yes
Institutional backing	OpenAI	Microsoft	Standalone	Standalone	Standalone

The table reveals a clear pattern. OpenClaw wins on every "real-world agent" capability: persistence, local execution, messaging integration, and proactive scheduling. The established frameworks win on documentation maturity and enterprise support structures.

For most teams evaluating agentic AI tooling right now, the trade-off is between bleeding-edge capability (OpenClaw) and better-documented stability (everyone else). That gap will close. The question is which direction it closes from.

The Security Paradox

It would be dishonest to write about OpenClaw's rise without discussing what the growth exposed.

By February 2026, security researchers had identified more than 135,000 publicly exposed OpenClaw instances. Approximately 15,000 were vulnerable to remote code execution via CVE-2026-25253. A campaign later named "ClawHavoc" placed over 800 malicious skills in the ClawHub marketplace, some delivering the Atomic macOS Stealer trojan to users who installed them without scrutiny.

Cisco Talos demonstrated that third-party skills could perform data exfiltration and prompt injection without triggering any visible warning to the user. Microsoft, CrowdStrike, 1Password, and Kaspersky each published separate guidance documents for organizations dealing with employee-installed OpenClaw agents on work devices.

This is not an argument against OpenClaw as a product. It is an argument about what the word "production-ready" actually means for agentic AI. Developer tooling optimizes for speed and capability. Enterprise tooling must optimize for governance, auditability, and containment as well. The 135,000 exposed instances are not a failure of the software; they are a failure of deployment practice, and they signal a gap that the industry needs to close deliberately.

What the Stars Actually Tell Us

A few things worth noting beneath the headline number.

Stars measure interest, not active usage. A significant portion of those 335,000 stars represent developers who bookmarked the project to evaluate later. The active user base is smaller, though still substantial given the fork count of over 65,000.

The fork-to-star ratio reveals depth of engagement. OpenClaw's ratio is approximately 1:5.1 (one fork for every 5.1 stars). AutoGen's ratio is roughly 1:9.75. A lower ratio means more watchers relative to active builders. OpenClaw has a broader, more casual audience and a narrower core of deep contributors, at least for now. That balance will shift as the project matures.

China is driving disproportionate adoption. Reports indicate that Chinese usage of OpenClaw matches or exceeds usage in the United States. Chinese developers refer to the project as "raising lobsters," a reference to the red claw logo. This signals genuinely global appetite for local AI agent infrastructure, not just a Western developer trend.

The Nvidia endorsement changed the audience. When Jensen Huang called OpenClaw "the next ChatGPT" at a public event, he was not speaking to developers. He was speaking to CTOs, CIOs, and enterprise investors. That endorsement triggered a qualitatively different category of attention: organizations evaluating strategic AI adoption, not just developers building weekend projects.

From Open-Source Excitement to Enterprise Deployment

If you are a CTO or engineering leader reading this, the instinct to pay close attention to OpenClaw is correct. The growth signal is real. The underlying demand, for AI agents that can take action across tools, systems, and communication channels with minimal human interaction, is a genuine shift in how software is going to be built and operated.

But the gap between "this is exciting developer tooling" and "this is safe to run in my organization" is also real.

OpenClaw is local-first, but 135,000 exposed instances show that local-first is not the same as secure-by-default. It has a skill marketplace, but that marketplace has been used to deliver malware. It supports multi-agent workflows, but without governance controls, a compromised agent has the same permissions as a trusted one. The very feature that makes it powerful, ambient access to files, APIs, and credentials, is what makes a misconfigured deployment a serious incident.

The agentic AI paradigm that OpenClaw made famous is not going away. The question for enterprise teams is not whether to adopt it; it is how to adopt it with the controls that the business actually requires: human-in-the-loop approvals for consequential actions, air-gapped deployment for sensitive workloads, audit trails that satisfy compliance teams, and access governance that does not depend on end-user configuration.

That is the engineering work that sits between an exciting open-source project and a production system.

At lowtouch.ai, we build exactly that layer: private-by-architecture agentic workflows with human-in-the-loop controls, SOC 2 Type II and ISO 27001 certification, and no-code configuration for teams that cannot afford to staff a full agent engineering program. If your organization is working through how to move from "OpenClaw is interesting" to "we have a governed AI agent program," we are worth a conversation.

Schedule a Conversation

What Comes Next

OpenClaw will not maintain this growth rate indefinitely. No project does. But the category it represents, always-on, locally-executed, messaging-native AI agents, is clearly where a large portion of developer energy is going in 2026.

The projects that follow will compete on trust, security, and enterprise readiness as much as on raw capability. The winner in three years may not be the most starred project today; it will be the one that figured out how to be genuinely safe to run in organizations where the stakes are high and the compliance requirements are real.

OpenClaw showed the world what demand looks like. The rest of the industry is now in a race to serve that demand at scale and safely.

335,000 stars in 60 days. That is the signal. How your organization responds to it is the strategy.