🦞 OpenClaw

From Weekend Project to 199K Stars

The Rise of the AI Agent Platform — February 2026

The Origin Story

Nov 2025

Peter Steinberger builds "Clawdbot" as a weekend project — a personal AI agent on WhatsApp

Jan 2026

Anthropic sends trademark complaint → renamed to "Moltbot" in a chaotic 5am Discord brainstorm

Late Jan

🚀 60,000 GitHub stars in 72 hours — 2M website visitors in one week

Feb 14

Peter Steinberger joins OpenAI. Project moves to independent foundation.

Feb 22

Final rebrand → OpenClaw. Version 2026.2.22 ships.

Andrej Karpathy: "the most incredible sci-fi takeoff-adjacent thing"

By The Numbers

199K

GitHub Stars

35K

Forks

5,700+

Community Skills (ClawHub)

11,440+

Commits

1.5M

AI Agents on Moltbook

50+

Channel Integrations

One of the fastest-growing open-source projects in history

What Is OpenClaw?

A self-hosted AI agent that lives on your machine and connects to every messaging platform — WhatsApp, Telegram, Discord, Slack, iMessage, Signal, Teams, LINE, Twitch, and more.

Text it a task → it does the task. That's it.

Model-agnostic — Claude, GPT, Gemini, DeepSeek, Mistral, Ollama, KIMI, MiniMax
Full data sovereignty — runs on your hardware, your data stays local
Always-on — 24/7 agent that remembers everything
MIT licensed — free, open source, no vendor lock-in

The OpenAI Hire

Sam Altman — "The future is going to be extremely multi-agent" — these capabilities will "quickly become core to our product offerings"

Peter Steinberger — "I'm a builder at heart. What I want is to change the world, not build a large company, and teaming up with OpenAI is the fastest way to bring this to everyone."

The bidding war: Steinberger chose OpenAI over Anthropic and Meta. Mark Zuckerberg personally reached out via WhatsApp. OpenClaw moves to an independent foundation with OpenAI financial backing.

Source: Fortune, Feb 17 2026

🤖 Sub-Agent Orchestration

v2026.2.15 — Agents spawning agents

Agents can now spawn child agents, which can spawn their own children. Configurable nesting depth, per-agent limits, depth-aware tool policies.

Depth 2+

Configurable Nesting

Max Children Per Agent

Break down complex tasks → spawn parallel workers → QC results → deliver. The multi-agent future Altman talked about.

🎮 Discord Components v2

v2026.2.15 — Discord becomes a control panel

🔘

Buttons

Clickable actions with primary/secondary/danger styles

📋

Select Menus

Dropdown selects for string, user, role, channel

📝

Modals

Pop-up forms with text inputs and paragraphs

📎

File Blocks

Attachment-backed file sharing in rich layouts

Not just text anymore — full interactive UI inside Discord

⌚ Apple Watch + iOS

v2026.2.19 — Your AI on your wrist

Watch inbox UI — read and respond to your agent from your wrist
Notification relay — agent alerts pushed to Watch
APNs wake — wake sleeping iOS devices via Apple Push Notifications
Remote invocation — trigger agent tasks from anywhere

Your AI assistant is now literally on your wrist. Ask it something while walking, get a response before you reach the door.

🌐 Browser Relay

Chrome Extension — Control your actual browser

A Chrome extension that lets the agent control your existing browser tabs — with your logins, cookies, and sessions intact. No headless browser needed.

Click the toolbar icon → agent takes over that tab
Navigate, click, fill forms, extract data — all automated
SSRF-guarded navigation for security
Auto-reconnect on connection drops

@Matt2Build — "My agent has been running on relay through chrome tabs for weeks. Yesterday the relay had an error, and claw just goes 'do it this way instead'."

🧩 Skills System + ClawHub

5,700+ community-built skills

5,700+

Skills on ClawHub

1 cmd

Install Any Skill

Self-Create

Agent Builds Its Own Skills

Weather, crypto analysis, image gen, SSH, presentations, research...
Agent can autonomously create new skills from scratch
Skills are just markdown + scripts — easy to build and share

🧠 Persistent Memory

v2026.2.22 — Now multilingual

Markdown-based local memory that persists across sessions. Daily files for context, MEMORY.md for long-term reference, optional vector search backend.

Daily memory — auto-creates memory/YYYY-MM-DD.md files
Long-term — MEMORY.md for permanent reference
Vector search — QMD backend for semantic recall
NEW: Multilingual — CJK, Arabic, Spanish, Portuguese, Japanese, Korean

Your agent wakes up fresh each session — these files ARE its memory

🔒 Security: The Elephant in the Room

CrowdStrike — Published a full security analysis + enterprise removal toolkit. Called it a potential "AI backdoor agent."

CVE-2026-25253 — Critical RCE vulnerability (CVSS 8.8/10) via WebSocket origin header exploitation

The response: 40+ security fixes per release. SSRF guards, sandbox hardening, prompt injection mitigations, rate limiting, auth token management, machine-checkable security models.

@minbeaky — "Strong release. The security hardening cadence is the right call — agent ecosystems need boring reliability more than flashy features."

📦 Latest: v2026.2.22

The "OpenClaw Rebrand" Release

Official rebrand — Moltbot → OpenClaw with full migration
New channels — Twitch + Google Chat plugins
New models — KIMI K2.5, Xiaomi MiMo-V2-Flash, Mistral AI
Multilingual memory — CJK/Arabic/Spanish/Portuguese/Japanese/Korean
Self-update — Agent updates itself with dry-run preview
34 security commits — continued hardening blitz

Community Reactions 🐦

@Legendaryy — "Best feature in the new OpenClaw update is the dry run. Means your agent can preview what's happening and adjust for any potential issues."

42 ❤️ • 5,315 views

@Dahilonx — "Watching OpenClaw blow up is a reminder: The best products don't add features. They remove friction. 'Text an agent → it does the task' is insane UX."

@HerrKarl6 — "I can now photograph chess positions, OpenClaw converts them, opens Firefox, sends it to an analysis tool, takes a screenshot of the best move and sends it back. Only €1.20 per query! 🥹"

OpenClaw vs Claude Code

Dimension	OpenClaw	Claude Code
Purpose	Life automation	Coding agent
Interface	50+ channels	Terminal / IDE
Models	Any (Claude, GPT, Gemini...)	Claude only
Data Control	Full sovereignty	Anthropic processes
Coding Ability	Basic	Superior (IDE integration)
Cost	Free + API ($5-30/mo)	$20-200/mo
Setup Time	30-60 minutes	30 seconds

They're complementary, not competitive. Claude Code is a surgical scalpel for coding. OpenClaw is a Swiss Army knife for life.

💻 The Mac Mini Trend

Developers are buying Mac Minis specifically to run OpenClaw 24/7 as a dedicated AI agent server.

@amit0xic (17K followers) — "finally picked up a Mac Mini to play around with openclaw this week. curious to see if the hype is real."

@jordymaui — "be proud of yourself for being a Mac Mini, there's a shortage"

DigitalOcean also launched a 1-Click OpenClaw Deploy ($24/mo) for those who don't want local hardware

🌱 The Ecosystem Exploding

🐾

PicoClaw

Go-based ultra-efficient alternative — <10MB RAM, <1s boot

💰

ClawAPI

API cost management layer for OpenClaw agents

☁️

Managed Hosting

Startups offering hosted OpenClaw at $3-5/month

📱

Moltbook

AI social network — 1.5M agents and counting

The Verdict

OpenClaw is the closest thing to JARVIS we've seen.

Weekend project → 199K stars → OpenAI acquisition in 3 months. The speed is unprecedented.

For power users: Sub-agents, browser control, 50+ integrations, full autonomy
For developers: MIT license, model-agnostic, 5,700+ skills, extensible
For everyone: Text your agent → it does the thing. That's the dream.

🦞

What feature are you most excited about?

Drop your thoughts in the comments

Sources: GitHub, Fortune, CrowdStrike, X/Twitter, OpenHands, docs.openclaw.ai