OpenClaw (formerly Moltbot, and Clawdbot) is a self-hosted gateway that turns your messaging apps (WhatsApp, Telegram, Discord, iMessage, Slack) into a powerful AI assistant. It runs on your own hardware, ensuring privacy and control.

Why so many names? Here is the breakdown:

• Origin: Inspired by Anthropic’s “Claude” models, but designed as a “bot” wrapper.
• Why it changed: To avoid trademark confusion with Anthropic’s official products. “Clawdbot” sounded too much like an official Claude bot.
• Relationship to Claude: OpenClaw is not affiliated with Anthropic, though it supports using Claude models (via API) as the brain of your assistant.

• Meaning: “Molt” refers to shedding skin and growing (like a crab or lobster). It symbolized the project’s evolution from a simple wrapper to a complex agent system.
• Why it changed: Users found the name confusing or hard to remember.

• Meaning: Emphasizes “Open” source and retains the “Claw” from the original crab/lobster theme (a nod to the Rust/Ferris ecosystem and the original name).

• Self-Hosted: You own your data. No middleman reading your chats.
• Multi-LLM Support: Connect to OpenAI, Anthropic, Gemini, DeepSeek, or local models (Ollama/LM Studio).
• Universal Gateway: One agent, available everywhere (WhatsApp, Telegram, Discord, Signal, iMessage, Slack, Matrix).
• Tool-Use: The agent can browse the web, write files, run code, and use custom “Skills”.

• Claude Code/Cursor: Focused on coding assistance within an IDE.
• OpenClaw: Focused on life assistance and automation via messaging apps. It lives in your phone’s chat app, not your code editor.

• ChatGPT/Gemini: Cloud-based, closed ecosystem. You can’t add custom tools easily, and they own the data.
• OpenClaw: Runs on your machine. You can give it access to your local files, your calendar, your smart home, and it can execute real shell commands.

• AutoGPT: Often autonomous “agents” that run in a loop to solve a task.
• OpenClaw: A “conversational” agent. It waits for you to talk to it, but can also run background tasks (cron jobs). It’s designed to be a long-term companion, not just a task solver.

You’re putting an agent in a position to:

• Run commands on your machine (exec tool).
• Read/write files in your workspace.
• Send messages back out via messaging apps.

Start Conservative:

1. Allowlist: Always set channels.<channel>.allowFrom to your personal user ID. Never run it open-to-the-world.
2. Sandbox: Use Docker sandboxing for tools if you are unsure.
3. Human in the Loop: Use exec.approvals to require confirmation before the agent runs dangerous commands.

Unlike cloud-based assistants, OpenClaw runs on your hardware (like a Mac Mini M4 or a Raspberry Pi), ensuring your conversations remain private and your tools have local access.