OpenClaw treats your messaging app as the interface instead of building its own chat window — sign in with an existing Anthropic, OpenAI, Google, xAI, or GitHub Copilot account, and it hands tasks off to whichever flagship model you already trust. Message it on WhatsApp, Telegram, Slack, Discord, iMessage, or one of roughly twenty other supported channels, and it replies in that same thread after browsing the web, running code, or reaching into other tools through its skill system.
It grew out of an earlier project called Warelay (later CLAWDIS, then Clawdbot, then Moltbot) before settling on the OpenClaw name in early 2026, and it's now developed in the open by the OpenClaw Foundation, a non-profit. The ecosystem around it has grown quickly too — the GitHub org hosts 87 repos, including a CLI for Google Workspace (gogcli), a macOS screenshot tool for visual agents (Peekaboo), and a TypeScript wrapper for calling MCP servers (mcporter). Pairing a phone as a connected device adds camera, screen capture, and voice control on top of the base chat interface.
The project's own docs are direct about the risk this creates: since anyone who can message the bot is effectively talking to an agent with tool access, OpenClaw's security guide says to treat every inbound DM as untrusted input. Source installs also need pnpm specifically — the docs warn that a plain `npm install` at the repo root isn't a supported setup, even though the published package installs fine through npm for regular use.