OpenClaw is an open-source, self-hosted personal AI agent framework that connects large language models (LLMs) to messaging platforms users already rely on — WhatsApp, Telegram, Slack, Discord, Signal, iMessage, and more. Originally launched as Clawdbot in November 2025 by Austrian developer Peter Steinberger (founder of PSPDFKit), it was briefly renamed Moltbot before settling on OpenClaw in January 2026. Written primarily in TypeScript, it became the fastest-growing project in GitHub history, surpassing React with over 335,000 stars by March 2026.

Unlike traditional chatbots confined to a browser tab, OpenClaw runs as a persistent daemon on the user's own hardware (macOS, Windows, Linux, or VPS) and acts as an autonomous “digital employee” — remembering context across conversations, executing shell commands, managing files, browsing the web, and proactively reaching out to users through their preferred channels.

Peter Steinberger published the original project as Clawdbot on November 24, 2025 — a weekend hack bridging Claude and Telegram. The name was a pun on Anthropic's Claude, but after a trademark request from Anthropic, the project was renamed to Moltbot (referencing how lobsters molt their shells to grow), then finally to OpenClaw on January 29, 2026. Within 72 hours of going viral, the repo cleared 60,000 GitHub stars. By early February 2026 it had crossed 185,000 stars, and by March 2026 it held over 335,000 stars with 360+ contributors and 65,000+ forks.

In February 2026, Steinberger joined OpenAI, and stewardship of the project transitioned to the OpenClaw Foundation at openclaw.org.

OpenClaw follows a hub-and-spoke architecture centered on the Gateway — a persistent WebSocket server process that routes messages between messaging channels and the AI agent core.

The Gateway is the central nervous system of an OpenClaw deployment. It maintains live, bidirectional WebSocket connections to all configured chat channels and routes every inbound message to the agent loop. It also manages session state, hooks, secrets, and tool invocations. A single Gateway process handles all channels — there is no separate per-channel service.

# Start the gateway
openclaw gateway --port 18789
 
# Verify health
openclaw gateway status
 
# Check channel connectivity
openclaw channels status --probe

Channels are adapters that connect external messaging services to the Gateway. Supported channels include WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Google Chat, and more. Each channel adapter translates platform-specific APIs into the internal OpenClaw message format.

The agent loop is the core reasoning engine. When a message arrives, the loop:

1. Loads user context and conversation history
2. Reads the mysoul.md canary file (defining user values and operational guidelines)
3. Sends the enriched prompt to the configured LLM
4. Parses the LLM response for tool calls or direct replies
5. Executes any requested tools/skills
6. Returns the result to the user via the originating channel

OpenClaw supports 14+ LLM providers out of the box, including OpenAI, Anthropic Claude, Google Gemini, DeepSeek, and local models via Ollama, vLLM, or LM Studio. Users can add custom providers through the models.providers configuration in openclaw.json.

OpenClaw uses a modular skill/plugin system with over 5,700 community-built skills. Skills are pluggable components that extend the agent's capabilities — from deploying websites (Vercel skill) to PDF analysis, web browsing, email management, and shell command execution.

Skills are registered in openclaw.json and can be allowlisted per agent:

// openclaw.json — enabling a plugin and allowlisting its tool
{
  "plugins": {
    "entries": {
      "llm-task": { "enabled": true }
    }
  },
  "agents": {
    "list": [
      {
        "id": "main",
        "tools": { "allow": ["llm-task", "pdf", "web-browse"] }
      }
    ]
  }
}

OpenClaw provides an interactive onboarding wizard:

# Install via npm
npm install -g openclaw
 
# Run the onboarding wizard (configures LLM provider, channels, and gateway)
openclaw onboard
 
# Or start directly with a specific provider
openclaw gateway --port 18789

The project also supports Docker deployment, local backups (openclaw backup create), and Git-backed rollback through community tools like AlphaClaw.

• Local-first: all data stays on the user's machine; no third-party data sharing required
• Multi-channel: single Gateway serves WhatsApp, Telegram, Discord, Slack, Signal, iMessage, and more
• Multi-model: swap between 14+ LLM providers (cloud or local) via config
• Persistent memory: remembers user context, preferences, and ongoing projects across sessions
• Autonomous execution: runs shell commands, manages files, browses the web, handles email
• Extensible skills: 5,700+ community-built skills; easy to create custom plugins
• Cross-platform: runs on macOS, Windows, Linux, and VPS environments
• MIT licensed: fully open-source and auditable

The OpenClaw ecosystem includes several notable companion projects:

• WebClaw — a fast web client for OpenClaw (https://github.com/ibelick/webclaw)
• AlphaClaw — setup harness with self-healing watchdog and browser-based observability (https://github.com/chrysb/alphaclaw)
• Clawith — multi-agent collaboration platform built on OpenClaw (https://github.com/dataelement/Clawith)
• LiteClaw — lightweight Python-based version of OpenClaw (https://github.com/Pr0fe5s0r/LiteClaw)

• OpenClaw GitHub Repository — 335,000+ stars, MIT license
• OpenClaw Official Site
• OpenClaw Documentation
• OpenClaw Foundation
• OpenClaw on Wikipedia
• Introducing OpenClaw — Official Blog Post
• Gateway Runbook — Official Docs

• OpenVoiceUI — Voice-driven AI interfaces
• Agent Orchestration — Multi-agent coordination patterns
• Function Calling — LLM function/tool calling mechanisms
• Model Context Protocol (MCP) — Anthropic's Model Context Protocol (MCP)