Running OpenClaw on Windows: native vs WSL2
OpenClaw runs on Windows natively (its Windows Hub app, or a Gateway as a Scheduled Task) and under WSL2, which the project calls the most Linux-compatible. How to choose.
Both paths work: native Windows (OpenClaw registers its Gateway as a Scheduled Task) and WSL2 (OpenClaw in a Linux environment inside Windows). The project now recommends its Windows Hub desktop app for a Windows PC and calls WSL2 the most Linux-compatible place to run the Gateway; since some skills and helper tools assume a Unix environment, WSL2 is the smoother choice for anything beyond a quick look, and a separate machine beats both once you’re serious.
Native Windows: real, with texture
The standard three-command install works on Windows with a supported Node version, and the docs also ship a native “Windows Hub” companion app (Windows 10 20H2+ and 11) as their recommended desktop experience, while still calling WSL2 the most Linux-compatible Gateway runtime on Windows. Where native Windows diverges is in the plumbing and the ecosystem:
- The daemon is a Scheduled Task. On macOS and Linux the Gateway runs under launchd/systemd; on native Windows,
openclaw gateway installregisters a Scheduled Task named “OpenClaw Gateway” (falling back to a Startup-folder launcher if task creation is denied). Two consequences people mistake for bugs: a terminal window can appear when the Gateway starts, and the agent keeps starting on boot until you stop it properly withopenclaw gateway stop(see the uninstall guide for making that permanent). - The ecosystem assumes Unix at the edges. Core OpenClaw runs fine, but skills that shell out to command-line tools often assume Linux/macOS conventions. That is why the docs point to WSL2 for the most Linux-compatible Gateway. If a skill misbehaves on Windows, this is the first suspect.
WSL2: Linux where the agent can feel at home
WSL2 (Windows Subsystem for Linux, installed with a single wsl --install command and a reboot) gives you a real Linux environment inside Windows, and OpenClaw inside it behaves like OpenClaw on Linux: systemd-style service management, Unix-convention tools, the environment most skills were written against. The cost is one layer of indirection: your agent’s state directory lives in the Linux filesystem (back that up, not just your Windows folders), and networking between Windows apps and the WSL2 VM occasionally needs a moment’s thought. For anyone planning to actually live with an agent on a Windows machine, WSL2 is the road with fewer surprises.
The question behind the question
Most people asking “can OpenClaw run on Windows?” are really asking “can I run it on the computer I already have?” The answer is yes, and that’s the right way to try it. But your Windows PC sleeps, reboots for updates on its own schedule, and holds your whole digital life inside the agent’s blast radius. Once the agent earns a permanent place, the Windows-vs-WSL2 question dissolves into a better one: which always-on machine should this actually live on? A small Linux server (or a managed pod, if you’d rather skip the server part) ends up being the answer for most people who keep their agent, with the Windows machine going back to being yours.