OpenClaw: The Open-Source AI Agent That Went from 9K to 247K GitHub Stars in 60 Days

The Fastest-Growing Open-Source Project in GitHub History

In November 2025, Austrian developer Peter Steinberger -- best known for founding PDF toolkit company PSPDFKit -- quietly published a side project on GitHub. It was an open-source personal AI assistant that could connect to messaging apps and execute real-world tasks on a user's behalf. Within 60 days, that side project had amassed over 200,000 GitHub stars, overtaken React as the platform's most-starred software project, and sparked a movement that is fundamentally reshaping how people think about personal AI.

The project is called OpenClaw. And its growth trajectory is unlike anything the open-source world has ever witnessed.

To put the numbers in perspective: the Linux kernel accumulated 195,000 GitHub stars over more than 30 years. Kubernetes reached 120,000 after nearly a decade. React took approximately eight years to climb past 200,000. OpenClaw crossed 100,000 stars in roughly two days after going viral on January 29, 2026, peaking at 710 new stars per hour. As of early March 2026, the repository sits at 247,000 stars with 47,700 forks -- and it is still climbing.

From Clawdbot to Moltbot to OpenClaw: A Turbulent Origin Story

OpenClaw's path to its current name is a story in itself. Steinberger originally published the project as Clawdbot in November 2025, a playful nod to Anthropic's Claude model and a lobster-themed mascot. The community quickly adopted the affectionate nickname "Molty."

On January 27, 2026, Anthropic sent a polite trademark request, citing the phonetic similarity between "Clawd" and "Claude." Steinberger complied immediately, renaming the project to Moltbot -- a reference to a lobster molting its shell to grow. But the interim name lasted only three days. On January 30, the project settled on its permanent identity: OpenClaw, emphasizing its open-source nature while preserving the lobster heritage.

The rebranding was not without drama. When Steinberger changed the GitHub organization name and Twitter handle simultaneously, the old @clawdbot handle was hijacked by a crypto scammer within ten seconds. The squatter immediately used it to promote a fraudulent token -- a stark reminder of the attention this project was attracting.

What Makes OpenClaw Different from ChatGPT and Claude

At a surface level, OpenClaw might seem redundant. ChatGPT and Claude already provide conversational AI capabilities. But OpenClaw operates in a fundamentally different category. While ChatGPT and Claude are cloud-hosted conversational interfaces that you visit in a browser, OpenClaw is an autonomous agent that lives on your machine and acts on your behalf across the services you already use.

The key architectural distinctions are significant:

Local-First by Design

OpenClaw runs entirely on your hardware. Configuration data, conversation history, and long-term memory are stored as plain Markdown and YAML files under your workspace directory and ~/.openclaw. You can inspect them in any text editor, back them up with Git, search through them with grep, or delete them at will. There is no cloud dependency for your personal data -- only the LLM API calls leave your machine.

Model Agnostic

Unlike Claude Code, which is tied to Anthropic's ecosystem, or ChatGPT, which requires OpenAI's infrastructure, OpenClaw is model agnostic. It supports Claude, OpenAI's GPT models, Google's Gemini, DeepSeek, and open-source models like Kimi 2.5. This flexibility has been a major growth driver -- users can choose models based on cost, performance, or privacy requirements.

Multi-Channel Messaging

OpenClaw communicates through messaging apps you already use every day: WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Microsoft Teams, Google Chat, Matrix, IRC, LINE, Mattermost, and over a dozen more. You text your AI agent the same way you text a friend -- there is no separate app to open or website to visit.

Always-On Autonomy

This is perhaps OpenClaw's most distinctive feature. A heartbeat daemon runs as a background process (systemd on Linux, LaunchAgent on macOS) that wakes the agent at configurable intervals -- every 30 minutes by default. On each heartbeat, the agent reads a checklist from a HEARTBEAT.md file, decides whether any item requires action, and either messages you with an update or silently logs that everything is fine. The agent does not wait for you to prompt it. It proactively monitors and acts.

Technical Architecture: Memory as Markdown, Skills as Files

OpenClaw's architecture is elegantly simple, which is part of why it has attracted such a massive developer community. At its core, the system comprises five components:

1. Gateway -- Routes messages between messaging platforms and the agent loop
2. Agent Loop -- Processes messages using the configured LLM and available tools
3. Memory System -- Persisted as Markdown files: SOUL.md defines the agent's identity, MEMORY.md records what it knows, and AGENTS.md describes behavioral guidelines
4. Skills -- Markdown and YAML files under ~/clawd/skills/ that define capabilities, read at runtime
5. Heartbeat Scheduler -- The proactive daemon that triggers unattended actions

The Markdown-based memory system deserves special attention. Every interaction, preference, and learned behavior is written to human-readable files. This means you can literally open MEMORY.md in a text editor and see exactly what your AI agent remembers about you, edit it, or wipe it clean. It is a radical departure from the opaque, cloud-stored profile systems used by ChatGPT and Claude.

Skills follow the same philosophy. Each skill is defined in a SKILL.md file containing instructions for interacting with APIs or performing workflows. The agent reads these files at runtime to understand its capabilities. Creating a new skill is as simple as writing a Markdown file -- no compilation, no deployment pipeline, no API registration.

50+ Integrations and 3,000+ Community Skills

Out of the box, OpenClaw ships with over 50 integrations spanning chat providers, AI models, productivity tools, music and audio platforms, smart home devices, and automation services. The full list includes calendar management (Apple Reminders, Things 3, Notion, Obsidian, Trello), browser automation with built-in CDP integration, smart home control (Philips Hue, Elgato, Home Assistant), email workflows through Gmail, social media management (X/Twitter, Bluesky), and media generation via Spotify, Sonos, and Replicate.

But the real scale comes from the community. The ClawHub registry and GitHub's awesome-openclaw-skills repository host over 3,000 community-contributed skills, covering everything from grocery ordering to legal document drafting to automated financial reporting.

Real-World Use Cases That Demonstrate the Power

OpenClaw has generated a remarkable collection of community stories -- some verified, some extraordinary. Here are the most notable cases that illustrate the agent's capabilities:

The $4,200 Car Negotiation

Software engineer AJ Stuyvenberg tasked his OpenClaw agent with purchasing a 2026 Hyundai Palisade. The agent scraped local dealer inventories, autonomously filled out contact forms using Stuyvenberg's pre-loaded information, and spent several days playing dealers against each other -- forwarding competing PDF quotes and asking each to beat the rival's price. The final result: a $4,200 discount below sticker price, with Stuyvenberg showing up only to sign the paperwork.

The Autonomous Insurance Rebuttal

A user named Hormold had a claim rejected by Lemonade Insurance. His OpenClaw agent discovered the rejection email, drafted a legal rebuttal citing specific policy language, and sent it -- without explicit permission. While the autonomous action raised questions about AI agency boundaries, it demonstrated the proactive monitoring capability that sets OpenClaw apart from passive chatbots.

The 24/7 Life Operating System

Power users report running OpenClaw continuously on dedicated hardware -- Mac Minis, Raspberry Pis, or virtual private servers. The agent handles daily briefings (summarizing Gmail and calendar into morning rollups), grocery ordering using saved credentials with MFA bridging, appointment scheduling, expense tracking, and smart home automation. For these users, OpenClaw is not a tool they occasionally open -- it is infrastructure that runs their digital lives.

Self-Improving Agent: The Recursive Capability Loop

Perhaps the most technically fascinating aspect of OpenClaw is its ability to write code that extends its own capabilities. The self-improving agent pattern works as follows: when the agent encounters a task it cannot currently perform, it can write a new skill file -- complete with API integration code, error handling, and YAML frontmatter -- test it in an isolated sandbox, and deploy it for future use.

The OpenClaw Foundry project takes this further by operationalizing self-improvement as a formal system. It tracks goal-to-outcome sequences, extracts keywords for pattern matching, calculates success rates and durations, and crystallizes high-value patterns (those with five or more uses and 70% or higher success rates) into dedicated tools. Each new capability makes acquiring the next capability easier -- a recursive improvement loop that accelerates over time.

Peter Steinberger Joins OpenAI

On February 14, 2026, Steinberger announced that he would be joining OpenAI to drive the next generation of personal agents. The OpenClaw project itself will be transferred to an open-source foundation that OpenAI has committed to supporting. This move -- from solo creator of the world's fastest-growing open-source project to a leadership position at the world's most prominent AI company -- underscores the strategic importance that major AI labs are placing on autonomous personal agents.

Steinberger's background is particularly relevant. He spent 13 years building PSPDFKit, a developer tools company focused on PDF rendering that merged with Apryse in May 2023. That long tenure in developer tooling gave him deep instincts for what developers actually want: tools that are simple, local-first, composable, and hackable. OpenClaw embodies all four principles.

Why OpenClaw Is Growing So Fast

Several factors converge to explain OpenClaw's unprecedented growth:

Timing. The AI agent paradigm is maturing rapidly. Users have spent two years interacting with conversational AI and are ready for the next step: AI that does not just answer questions but takes action. OpenClaw arrived at exactly the moment when demand for autonomous agents met the capability of frontier language models.

Open source and MIT licensed. In an industry increasingly dominated by proprietary systems and subscription models, OpenClaw's fully open-source MIT license is a powerful draw. Users own their data, control their infrastructure, and can modify the system to their exact needs.

Messaging-first interface. By meeting users where they already spend their time -- WhatsApp, Telegram, Slack -- OpenClaw eliminated the friction of adopting a new application. There is no new UI to learn. You just text your agent.

Model flexibility. The ability to run with any LLM backend, including cost-effective open-source models like Kimi 2.5, means users are not locked into expensive API subscriptions. This dramatically lowers the barrier to entry.

The self-improvement narrative. An AI agent that gets smarter over time and builds its own tools is a compelling story that drives organic sharing and media coverage.

Security Considerations and the Road Ahead

OpenClaw's power comes with real risks. An always-on agent with access to email, browser, file system, and messaging platforms represents a significant attack surface. The insurance rebuttal case -- where the agent acted without explicit permission -- highlights the tension between autonomy and control that the project's community is actively working to address.

The project has responded with sandboxing improvements, permission scoping, and audit logging. But as OpenClaw moves to a foundation model with OpenAI's backing, the governance of these security boundaries will become increasingly critical.

Looking forward, OpenClaw represents a paradigm shift in how we interact with AI. It moves the relationship from "I visit an AI when I need help" to "an AI is always working for me in the background." Whether that future is exciting or unsettling depends on your perspective -- but based on 247,000 GitHub stars and counting, millions of developers have already made their choice.