TL;DR: The openclaw/openclaw README is the front door to the open-source personal AI assistant that runs on your own device. This guide walks it section by section: the positioning line that explains what OpenClaw is, the three install paths (shell script for macOS/Linux/WSL2, PowerShell for Windows, npm for Node 22+ environments, plus Docker/Nix), the quick-start's three verification steps, and the architecture vocabulary — Gateway, Channels, Tools, Model Providers — that every other doc assumes you know. MIT-licensed, maintained by the OpenClaw Foundation.
What the README is actually telling you
Every README answers three questions: what is this, how do I run it, and where do I go deeper. The OpenClaw README does all three efficiently, but it assumes you can decode its vocabulary. Here is the decoded version.
The one-line positioning
The README opens by defining OpenClaw as a personal AI assistant that runs on your own device and meets you in the chat channels you already use. Three claims are packed in there:
- Personal — one agent per person, with persistent memory, not a stateless chatbot
- Your own device — the control plane runs locally; your data stays with you (why that matters)
- Channels you already use — WhatsApp, Telegram, Slack, Discord — no new app to install
The install section, decoded
The README offers several paths to the same outcome. Which one you want depends on your environment:
1. Quick install script (macOS / Linux / WSL2):
curl -fsSL https://openclaw.ai/install.sh | bash
The fastest route on Unix-like systems. It handles dependencies and gets you to onboarding in one command.
2. PowerShell (Windows): the equivalent one-liner for native Windows setups.
3. npm package: for environments where you manage Node yourself — the README pins a current Node 22+ requirement. Natural choice if you already live in the JavaScript toolchain.
4. Docker / Nix: containerized and reproducible setups for people with opinions about their infrastructure. Our Docker setup guide goes deeper.
If you would rather skip all of this, that is the exact gap OneClaw fills — managed hosting for the same open-source core. The honest trade-off is covered in OneClaw vs self-hosting OpenClaw.
Quick start: the three checks that matter
The README's quick start is a three-step verification loop: run onboarding (configure your model provider and first channel), check gateway status, and open the dashboard. If all three pass, you have a working agent. Most first-run problems trace back to skipping onboarding details — model credentials and channel tokens — rather than anything exotic.
The architecture vocabulary
Four terms from the README structure everything else in the docs:
- Gateway — the local control plane. It runs on your machine, owns state, and coordinates every other piece.
- Channels — the messaging integrations (WhatsApp, Telegram, Slack, Discord and more). One agent, reachable everywhere you chat. Setup walkthrough: OpenClaw Telegram guide.
- Tools / Skills / Plugins — the extension surface: what your agent can actually do beyond conversation.
- Model Providers — the intelligence layer, pluggable between hosted APIs and local models. Cost implications in our hosting and model guide.
This modularity is the reason the project scales from a laptop hobby setup to serious self-hosted deployments without changing its shape.
The links most people skip (and should not)
The README links out to Configuration, Architecture, Updating, and — most importantly — Security and Sandboxing guides. An agent that can read your messages and execute tools deserves deliberate isolation choices. If you deploy beyond your laptop, read those two before exposing anything — and see our firewall/VPN deployment guide for a hardened pattern.
Governance: who maintains this
OpenClaw is MIT-licensed and maintained by the OpenClaw Foundation, a non-profit, with community-driven development in a pnpm-workspace monorepo. Practically, that means: free to use commercially, fork-friendly, and not hostage to a single vendor's pricing decisions.
Where to go next
- Want the repo tour beyond the README? OpenClaw GitHub: repo guide and 10-minute setup
- Ready to deploy without the DevOps? Deploy a managed OpenClaw agent with OneClaw — same core, one click
- Comparing paths first? OneClaw vs self-hosting lays out the real costs of each.