Posted by The Buzzbaba Automation Team

OpenClaw has rapidly evolved from a niche open-source project into a widely used local AI agent runtime. Unlike standard chatbots, OpenClaw runs directly on your machine, capable of executing shell commands, managing files, and integrating with apps like Discord, WhatsApp, and Telegram.
For Windows users, getting a local autonomous agent running requires specific configuration, particularly regarding the Windows Subsystem for Linux (WSL).
This guide walks you through setting up OpenClaw (Moltbot) on Windows 10 or 11, ensuring a stable environment for your personal AI agent.
What you will learn
- How to prepare your Windows environment with WSL2.
- Installing the necessary dependencies (Node.js).
- Running the OpenClaw installer and onboarding wizard.
- Troubleshooting common gateway connectivity issues.
Prerequisites
Before installing the OpenClaw CLI, ensure your system meets the following requirements. OpenClaw relies on a Unix-like environment to execute many of its autonomous functions effectively.
- OS: Windows 10 (Version 2004+) or Windows 11.
- Environment: WSL2 (Windows Subsystem for Linux) is highly recommended over raw PowerShell for stability.
- RAM: Minimum 4GB dedicated to the agent process (8GB+ recommended if running local LLMs).
Step 1: Set up WSL2
OpenClaw functions best in a Linux environment due to its reliance on system-level process management. We will use Ubuntu via WSL.
- Open PowerShell as Administrator.
- Run the following command to install WSL and Ubuntu:PowerShell
wsl --install - Restart your computer if prompted.
- After rebooting, open the “Ubuntu” app from your Start menu. You will be asked to create a UNIX username and password.
Note: If you already have WSL installed, ensure you are running version 2 by executing
wsl --set-version Ubuntu 2in PowerShell.
Step 2: Install Node.js
OpenClaw is built on Node.js. The agent requires Node.js 22 or higher to handle the latest WebSocket and AI streaming protocols.
Run the following commands inside your Ubuntu terminal (not PowerShell):
- Update your package lists:Bash
sudo apt update && sudo apt upgrade -y - Install Node.js 22:We recommend using the official NodeSource repository for the latest version.Bash
curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash - sudo apt-get install -y nodejs - Verify the installation:Bash
node -v # Output should be v22.x.x or higher
Step 3: Install OpenClaw
With the environment ready, you can now install the OpenClaw core.
Option A: The Quick Script (Recommended)
The easiest way to install OpenClaw and its dependencies is via the official install script. In your Ubuntu terminal, run:
Bash
curl -fsSL https://openclaw.ai/install.sh | bash
This script detects your OS, installs the global NPM package, and prepares the background daemon.
Option B: Manual NPM Install
If you prefer more control over the installation:
Bash
npm install -g openclaw@latest
Once the package is installed, initialize the daemon:
Bash
openclaw onboard --install-daemon
Step 4: Configuration and “Hatching”
After installation, the “onboarding” process initializes your agent. This is where you connect your AI models and chat interfaces.
- Select a Model Provider:The wizard will ask for your preferred LLM.
- Cloud: OpenAI (GPT-4o), Anthropic (Claude 3.5 Sonnet).
- Local: Ollama (DeepSeek-R1, Llama 3).
- Tip: For local installs, we recommend starting with a cloud provider like Anthropic for better reasoning before switching to local models.
- Connect a Interface:You can connect OpenClaw to Telegram, Discord, or simply use the Web UI.
- If choosing Telegram, you will need a bot token from
@BotFather. - If skipping, you can access your agent via the local dashboard.
- If choosing Telegram, you will need a bot token from
- Hatch Your Agent:Give your agent a name. The system will generate a
config.jsonfile in~/.openclaw/.
Step 5: Accessing the Dashboard
Once the daemon is running, OpenClaw serves a local web interface for monitoring tasks and managing skills.
- URL:
http://localhost:18789 - Default Login: Setup during onboarding.
From here, you can install “Skills” (plugins) that allow your agent to read emails, manage your calendar, or browse the web.
Troubleshooting
If your agent isn’t responding, check these common Windows/WSL issues.
Issue: “Gateway not responding”
This often happens if systemd is not active in WSL. OpenClaw needs systemd to keep the agent running in the background.
Fix:
- Edit the WSL config:
sudo nano /etc/wsl.conf - Add these lines:Ini, TOML
[boot] systemd=true - Save and exit. Restart WSL by running
wsl --shutdownin PowerShell, then reopen Ubuntu.
Issue: Port 18789 in use
If the dashboard won’t load, another service might be blocking the port.
Fix:
Run sudo lsof -i :18789 to identify the process, or change the port in ~/.openclaw/config.json.
Next steps
You now have a fully functional AI agent running locally on Windows.
To get the most out of OpenClaw, try installing the “Browser Automation” skill from the dashboard. This allows your agent to research topics and provide summaries directly in your chat interface.
Ready to build your own skills? Check out our Developer Documentation to learn how to write custom TypeScript functions for your agent.
Frequently Asked Questions regarding OpenClaw Installation
Why does Windows Defender flag OpenClaw as a virus?
This is a known false positive. Because OpenClaw is an autonomous agent that can “execute shell commands” and “read files,” Windows Defender’s heuristics often flag the installer (specifically the `node.exe` wrapper) as potential malware. To fix this, you must add the installation folder ~/.openclaw to your exclusion list or temporarily disable Real-Time Protection during install.
I get “Error 0xc0000005” when running the installer. How do I fix it?
Error 0xc0000005 is an Access Violation error, typically caused by permissions issues in PowerShell. To resolve this:
- Right-click the downloaded installer and check “Unblock” in Properties.
- Run PowerShell as Administrator.
- If using WSL, ensure you are not trying to run the Windows binary from inside Linux. Use the Linux installation script instead.
Why can’t I access the dashboard at localhost:18789?
If you are running OpenClaw inside WSL2, the port might not be forwarded to Windows correctly. First, try accessing it via your WSL IP address (run ip addr to find it). If that fails, the port may be blocked by the Windows “IP Helper” service. Run wsl --shutdown and restart to clear the network stack.
How do I switch from OpenAI to local Ollama models?
Open your config file located at ~/.openclaw/config.json. Change the "provider" field from "openai" to "ollama" and set the "model" to your downloaded model (e.g., "deepseek-r1" or "llama3"). Restart the agent daemon for changes to take effect.
Does OpenClaw work on Windows 10 Home?
Yes, but only if you have WSL2 enabled. The legacy WSL1 does not support the background daemons (systemd) that OpenClaw requires. Ensure you are on Windows 10 version 2004 or higher.
What is “Moltbook” and why is my agent posting there?
Moltbook is the decentralized “social network” for OpenClaw agents. By default, the "social_mode" is set to true in the config, allowing your agent to “talk” to other agents to learn new skills. You can disable this by setting "social_mode": false in your configuration file.
Where is the config.json file located on Windows?
If you installed via PowerShell, it is at C:\Users\YourName\.openclaw\config.json. If you installed via WSL (Ubuntu), it is located at /home/username/.openclaw/config.json inside the Linux file system.
Can OpenClaw read my local files?
Yes, but only the directories you explicitly whitelist. During the onboarding wizard, you are asked to provide a “Safe Zone” path. The agent cannot access files outside of this directory unless you run it with the --unsafe flag.
I get “Gateway Not Responding”. What do I do?
This usually means the background service crashed. Run openclaw status to check the logs. If you see a “Memory Limit Exceeded” error, you may need to increase the memory allocation in your .wslconfig file.
Is OpenClaw free to use?
The OpenClaw software is open-source (MIT License) and free. However, if you use cloud models like GPT-4o or Claude 3.5, you must pay those providers directly via API keys. Using local models via Ollama is completely free.


Leave a Reply