OpenClaw Setup Guide (2026): Install Your Personal AI Assistant and Get the Most From It

Note on naming: the project is spelled OpenClaw (one word). If you searched for "open claw," you are in the right place.

What OpenClaw Is (In Plain Terms)

OpenClaw is a multi-channel AI gateway plus assistant runtime that you operate on your own machine (or server). The gateway is the control plane; you chat through a browser dashboard and optionally through messaging channels such as Telegram, Slack, Discord, and many others described in the Channels documentation. The project website is openclaw.ai.

What You Need Before You Start

1. Node.js: The docs recommend Node 24 and also support Node 22.14+. Check with node --version. If you need install help, see the Node setup page.
2. Model access: You will need credentials for a supported model provider (for example Anthropic, OpenAI, or others). The onboarding wizard walks you through auth—API keys, OAuth, or provider-specific flows depending on what you pick. See Models in the docs for selection and configuration.
3. Windows users: OpenClaw supports native Windows and WSL2. The official platform guide states that WSL2 is more stable and recommended for the full experience. Native Windows works for core CLI and gateway scenarios but has documented caveats (for example around scheduled tasks and non-interactive onboarding). Read Windows before you commit to a path.

Step 1: Install OpenClaw

The recommended path in the getting-started guide uses the official installers.

macOS or Linux

Run the install script from a terminal:

curl -fsSL https://openclaw.ai/install.sh | bash

Windows (PowerShell)

iwr -useb https://openclaw.ai/install.ps1 | iex

Alternative: npm (global)

The npm package documents npm install -g openclaw@latest (or pnpm). The full install menu—including Docker and Nix—is on the Install page.

Step 2: Run Onboarding (Recommended)

CLI onboarding is the recommended way to configure a local gateway (or a remote gateway connection), channels, skills, and workspace defaults in one flow. From the Onboarding (CLI) documentation, the entry command is:

openclaw onboard

To install the gateway as a background service (daemon)—LaunchAgent on macOS, systemd user unit on Linux/WSL2, or Windows Scheduled Task with Startup-folder fallback on native Windows—use:

openclaw onboard --install-daemon

What the wizard covers

According to the official wizard reference, local onboarding typically includes:

1. Model and auth — provider, keys or OAuth, default model. The docs warn that weaker models are easier to prompt-inject when tools or webhooks are involved; prefer the strongest model you can run with a strict tool policy.
2. Workspace — where agent files live (default ~/.openclaw/workspace).
3. Gateway — port, bind address, auth mode. QuickStart defaults include port 18789 and a generated gateway token.
4. Channels — optional integrations (Telegram, WhatsApp, Slack, and others).
5. Daemon — installs the service so the gateway survives reboots where supported.
6. Health check — starts the gateway and verifies it is running.
7. Skills — recommended skills and optional dependencies.

You can choose QuickStart (defaults) or Advanced (full control). There is also a web search configuration step with optional providers (some need API keys). Re-running onboarding does not wipe your setup unless you explicitly reset; if the config is invalid, the docs suggest running openclaw doctor first.

Step 3: Verify the Gateway

openclaw gateway status

You should see the gateway listening on port 18789 (unless you changed it). Open the control UI:

openclaw dashboard

If the dashboard loads, send a test message in the Control UI chat. The docs note this is the fastest first chat path—no channel setup required. Dashboard details: Dashboard.

How to Get the Most Out of OpenClaw

1. Add one messaging channel you actually use

The getting-started guide highlights Telegram as a fast setup option (bot token). Pick one channel, finish pairing, and validate end-to-end before you wire up everything. Always read Channels and the pairing and safety documentation so you control who can talk to your assistant.

2. Treat the gateway like production infrastructure

• Prefer strong models and conservative tool policies for anything exposed to untrusted messages or webhooks.
• Store secrets in environment variables or supported secret references—not in chat logs or random config snippets.
• Run openclaw doctor when something looks wrong or after upgrades. See Updating.

3. Reconfigure without starting from scratch

Use openclaw configure to adjust sections later (for example web search via openclaw configure --section web). Add agents with openclaw agents add <name> when you need separate workspaces or routing.

4. Explore tools and extensions deliberately

The docs index covers browser automation, exec, web search, skills, plugins, and MCP-style integrations. Add one capability at a time, test it, and review permissions. Jumping straight to "everything enabled" is how you get surprising file or network access.

Troubleshooting (High-Signal)

• Gateway not listening: Confirm openclaw gateway status, check port conflicts, and review the Gateway runbook.
• Node version errors: Upgrade to Node 24 or at least 22.14+ per docs.
• Windows oddities: If native Windows service install fails, read the Scheduled Task and fallback notes on the Windows platform page; consider WSL2 for parity with Linux behavior.
• Invalid or legacy config: Run openclaw doctor and follow its guidance before re-onboarding.

Summary Checklist

1. Install Node and confirm version.
2. Run the official install script (or npm global / other documented method).
3. Run openclaw onboard --install-daemon.
4. Verify with openclaw gateway status and openclaw dashboard.
5. Chat in the Control UI, then add one channel with pairing rules you understand.
6. Use openclaw doctor when upgrading or debugging.

Conclusion

OpenClaw gives you a serious, self-hosted assistant stack, but the winning move is boring: correct Node version, clean onboarding, verified gateway, one channel at a time, and conservative permissions. Stay pointed at the official docs as the product evolves.