What is OpenClaw?
A comprehensive introduction to OpenClaw, the open-source AI agent framework with 330K+ GitHub stars
OpenClaw is a free, open-source AI agent framework that runs on your machine and connects large language models to your apps, devices, and services. Unlike simple chatbots that only generate text, OpenClaw agents take real-world actions — reading and writing files, running shell commands, browsing the web, sending emails, controlling smart home devices, and more.
It is the fastest-growing open-source project in GitHub history, surpassing 331,000 stars by March 2026 (overtaking React). Its mascot is a lobster. Its tagline: "The lobster way."
Background and History
OpenClaw was created by Peter Steinberger, an Austrian developer best known for founding PSPDFKit. He first released the project in November 2025 under the name "Clawdbot." After a trademark issue with Anthropic, the project was briefly renamed to "Moltbot" on January 27, 2026, then settled on "OpenClaw" three days later on January 30.
When Steinberger joined OpenAI on February 14, 2026, the project was transferred to an open-source foundation to ensure its independence. OpenClaw is MIT-licensed and written in TypeScript.
The project gained 60,000 GitHub stars in its first 72 hours — a record that still stands.
How It Works
OpenClaw uses a Gateway architecture. The Gateway is an always-on process (default port 18789) that manages sessions, channels, tools, and memory. It acts as the central hub between your messaging channels and the LLM.
The agent loop works like this:
- A message arrives from a channel (Telegram, Discord, console, etc.)
- The Gateway routes it to a session and builds context (conversation history, persistent memory, relevant skills)
- The LLM receives the context and decides which tools to call
- Tools execute inside a sandbox (using bubblewrap/bwrap for safe command execution)
- The response flows back through the channel
All configuration lives in a single openclaw.json file (JSON5 format), making it straightforward to version-control and replicate your agent setup.
The Gateway is a persistent process, not a request-response server. It maintains WebSocket connections, manages conversation state, and keeps your agent "alive" between messages.
What It Can Do
OpenClaw agents are general-purpose. Here are real-world examples from the community:
Personal productivity — Morning briefing agents that summarize email, calendar, and news. Users have reported clearing backlogs of 4,000+ emails with agent-assisted triage.
Development and DevOps — Code review agents, multi-agent orchestration for complex engineering tasks, CI/CD pipeline monitoring, and automated issue triage.
Business automation — Solo founders running multi-agent operations have reported generating $62K+ in revenue with OpenClaw handling customer communication, content, and workflows.
Smart home — Control Philips Hue lights, Sonos speakers, and Home Assistant devices through natural language via any connected messaging channel.
Content and social media — Automated publishing pipelines that draft, review, and post content across platforms.
The Ecosystem
Models
OpenClaw is model-agnostic. It supports 20+ LLM providers out of the box, including Claude, GPT, Gemini, DeepSeek, Grok, Qwen, and local models via Ollama. About 80% of supported providers use the OpenAI-compatible API format, so adding a new provider typically requires only a configuration change (base URL and API key) — no code.
The three core protocol adapters are:
- openai-completions — covers most providers
- anthropic-messages — for Claude models
- google-generative-ai — for Gemini models
Channels
OpenClaw connects to 15+ messaging platforms: WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Microsoft Teams, Google Chat, Matrix, IRC, and more. Each channel runs independently, so a single agent can serve users across multiple platforms simultaneously.
Skills
The ClawHub marketplace hosts over 13,700 community-contributed skills, with 53 skills bundled in the default installation. Skills extend what your agent can do — from web search and file management to domain-specific workflows. OpenClaw also supports MCP (Model Context Protocol) for tool extensibility.
Skills are composable. You can combine multiple skills to create sophisticated workflows — for example, pairing a web-scraping skill with a summarization skill and a Slack-posting skill to build an automated research pipeline.
Getting Started
The standard way to run OpenClaw is self-hosted: install it on your own machine or server, configure your LLM API keys, and start the Gateway. The project has thorough documentation and an active community on GitHub and Discord.
If you'd rather skip the setup and get an OpenClaw agent running in 30 seconds, Claw Buff handles deployment, security, and updates for you.