# OpenClaw – Complete Guide, Part 1: Installation and First Launch > Part one of the practical OpenClaw guide: requirements, local vs VPS installation, the onboarding wizard, and daemon configuration. - URL: https://mautoblog.com/en/posts/openclaw-complete-guide-installation-part-1/ - Language: en - Published: 2026-02-25 - Updated: 2026-10-13 - Author: Mauto - Categories: Artificial Intelligence, Software Development - Tags: OpenClaw, Self-Hosted, JavaScript, AI Agents - Guide: OpenClaw from A to Z (part 1) - Translation (it): https://mautoblog.com/posts/openclaw-guida-01-installazione/ --- You've read the introductory post on OpenClaw and decided to give it a try. Good. In this first part of the guide we'll install the gateway, choose where to run it, and get it started for the first time. No theory: just concrete steps. **What you'll learn in this article:** - The minimum requirements to run OpenClaw - The difference between local and VPS installation (and which to choose) - How to install OpenClaw and complete the initial onboarding wizard - How to keep it running as a background service --- ## Prerequisites: What You Need Before Starting Before touching any command, make sure you have: **Required software** - **Node.js version 22 or higher** — OpenClaw runs on Node. If you don't have it, download it from `nodejs.org` or use `nvm` to manage versions. - **At least one API key for a language model** — The most recommended one to start is Anthropic (Claude). You'll find it at `console.anthropic.com`. You can also use OpenAI, Google Gemini or local models via Ollama. - **npm** — Installed automatically alongside Node.js. **Minimum knowledge required** - Knowing how to open a terminal and type commands - Knowing how to edit a text file (even with `nano` or any editor) - Being familiar with the concept of environment variables and JSON configuration files If these concepts are new to you, take an hour for a basic terminal tutorial before continuing. It's worth the investment. --- ## First Decision: Local or VPS? This is the most important choice you'll make. It has implications for security, cost, and convenience. ### Local installation (on your computer) **Advantages:** Immediate setup, no additional cost, easy to experiment with. **Disadvantages:** OpenClaw only works when your computer is on. If you shut down your Mac or PC, the agent stops working. Not suitable for proactive 24/7 tasks. **When it makes sense:** Testing and learning phase. If you just want to understand how it works before investing in a server. ### VPS installation (cloud server) **Advantages:** The agent runs 24/7, 7 days a week, even while you sleep. Your personal computer stays isolated and safe. This is the recommended mode for serious use. **Disadvantages:** Has a monthly cost (an entry-level VPS with 2GB RAM costs between €4 and €10/month on providers like Hetzner, DigitalOcean or Vultr). Requires a few extra steps for the initial configuration. **When it makes sense:** As soon as you decide to use OpenClaw for something more than an experiment. The separation between the agent's server and your personal machine is also a fundamental security measure. > **Recommendation:** Start locally to get familiar. Then, when you're ready, migrate to VPS. This guide covers both scenarios. --- ## Installation ### Method 1: One-liner script (the fastest) Open your terminal and run: ```bash curl -fsSL https://openclaw.ai/install.sh | bash ``` The script downloads OpenClaw, configures the environment, and prepares you for the next step. It's the fastest method and handles dependencies automatically. > **Security note:** Before running any script downloaded from the internet with `curl | bash`, it's good practice to visit the script URL in your browser to read its contents. It's one extra step, but it's a good habit. ### Method 2: Installation via npm (more controlled) If you prefer a more direct approach: ```bash npm install -g openclaw ``` The `-g` option installs the package globally, making it available as a command from any folder on the system. Verify the installation was successful: ```bash openclaw --version ``` You should see the current version number. If the command is not recognized, you may need to restart the terminal or add the npm global packages folder to your `PATH`. --- ## First Launch: The Onboarding Wizard OpenClaw includes an interactive wizard that guides you through the initial configuration. Launch it with: ```bash openclaw onboard ``` The wizard will ask you a few fundamental questions. Here's what to expect: **1. API key entry** You'll be asked to enter the API key for your preferred AI provider. For Anthropic (Claude), you'll find it in your dashboard at `console.anthropic.com`. Paste it when prompted. **2. Default model choice** You can choose which model to use by default for the agent. To start, `claude-haiku` or `claude-sonnet` are great choices: they offer a good balance between capability and cost. Avoid setting `claude-opus` as the default model: it's powerful but expensive, and using it for all tasks is wasteful. **3. Messaging channel configuration** The wizard will offer to connect a channel right away (Telegram is the simplest to start with — we'll cover it in detail in article 2). You can skip this step for now and configure channels later. **4. Confirmation and first test** At the end, the wizard will briefly start the gateway to verify everything works. If you see the confirmation message, you're good to go. --- ## Installing as a Service (VPS or Continuous Use Only) If you want OpenClaw to stay active even after closing the SSH session (or the terminal), you need to install it as a **daemon** — a background process managed by the operating system. ```bash openclaw onboard --install-daemon ``` This command registers OpenClaw as a system service. From this point on, the gateway will start automatically on every server reboot, without you having to do anything. To check the service status: ```bash openclaw status ``` To stop or restart the gateway manually: ```bash openclaw stop openclaw restart ``` --- ## The Web Dashboard: Your Control Center Once started, OpenClaw exposes a web dashboard accessible from the browser. By default, it's located at: ``` http://127.0.0.1:18789/ ``` From here you can: - See active conversations with the agent - Manage connected channel sessions - Check logs in real time - Configure some settings without manually editing configuration files > **VPS warning:** If you're using a remote server, don't expose port 18789 on the internet without protection. Use SSH tunneling (`ssh -L 18789:localhost:18789 user@your-server`) to access the dashboard securely. In article 2 we'll look at more elegant solutions with Tailscale or Cloudflare. --- ## Where Are the Configuration Files All of OpenClaw's configuration lives in a single JSON file: `~/.openclaw/openclaw.json` You can open it with any text editor. For now there's no need to edit it manually — the wizard took care of the basic settings. In article 2 we'll see how to customize it for security and channel configuration. --- ## Summary and Next Step At this point you should have: - OpenClaw installed and running (locally or on VPS) - The onboarding wizard completed with at least one API key configured - The web dashboard accessible from the browser - Optionally: the daemon service active for 24/7 operation In the next article we'll connect OpenClaw to Telegram and WhatsApp, configure the allowlist for security, and see how to protect the gateway from unauthorized access. --- *Next article → [Part 2: Channel Configuration and Security](/posts/openclaw-complete-guide-channels-security-part-2/)*