OpenClaw is an open-source personal AI assistant Gateway you run on your own machines. The repo is github.com/openclaw/openclaw — MIT-licensed, written in TypeScript, with sponsors including OpenAI, GitHub, NVIDIA, Vercel, Blacksmith, and Convex. Catchphrase: "EXFOLIATE! EXFOLIATE!" (the project is lobster-themed 🦞).

You install it via npm install -g openclaw@latest && openclaw onboard --install-daemon. It runs as a launchd / systemd daemon and exposes a Gateway server (default port 18789, token auth) that bridges:

24+ messaging channels — WhatsApp, Telegram, Slack, Discord, Google Chat, Signal, iMessage, IRC, Microsoft Teams, Matrix, Feishu, LINE, Mattermost, Nostr, WeChat, QQ, WebChat, plus macOS / iOS / Android nodes.
Configurable AI agents — each agent has its own workspace at ~/.openclaw/agents/<agentId> with SOUL.md, AGENTS.md, USER.md, and a session store. Models are pluggable with fallbacks across Anthropic Claude, OpenAI GPT, Gemini, Groq, OpenRouter, and others.
Built-in tools — browser, canvas, nodes, cron, sessions, plus Discord/Slack action APIs. Voice Wake on macOS/iOS, continuous voice on Android.
Routing bindings — route a specific channel/peer to a specific agent (e.g. coder agent for slack; work agent for the engineering Telegram group).
Sandbox modes — off / non-main / all, with Docker / SSH / OpenShell backends. DM pairing default for unknown senders.

OpenClaw is the substrate. It doesn't pick your model for you, it doesn't host your data for you, and it doesn't manage Telegram bot tokens for you — you do all of that.

ShipClaw is the managed version: we run an OpenClaw Gateway for you, narrow the model to Anthropic Claude (for predictable unit economics), wire your Telegram BotFather token into the right channel config, ship a custom SOUL.md for your agent, and bill credits instead of asking you to bring an Anthropic key.

Adjacent ecosystem: ClawHub (the OpenClaw skills and plugins marketplace), ClawHost (a managed-OpenClaw vendor — see our comparison), plus Discord community.