Here’s the bottom line up front. In February 2026, we wrote an install guide on the premise that “OpenClaw doesn’t officially support Windows, so you need WSL2.” That premise no longer holds. But there’s a precise distinction worth making here — a native app shipping doesn’t mean the WSL2 technology itself disappeared. What disappeared is the 8-step manual ritual users used to walk through by hand; the tech underneath it is still very much in use. This piece draws that line precisely, and lays out the three install paths that actually exist today.
🦞 What OpenClaw Is, and Why We’re Rewriting This
OpenClaw is a locally-run, open-source AI agent gateway. It connects AI to messengers like Discord, Telegram, and WhatsApp, so you can chat naturally and hand off tasks. Peter Steinberger built it, and today he leads “Claw Labs” inside OpenAI, retaining final technical say, while the OpenClaw Foundation — a 501(c)(3) nonprofit formally launched on July 8, 2026 — now manages the project. That Foundation launch happened 12 days before we’re writing this piece.
The old post covered exactly one path, on the premise that “OpenClaw doesn’t officially support Windows.” That premise itself is now outdated. Below, we lay out the three install paths that actually exist today.
Which path fits depends on what you’re after. If you just want to get moving quickly and chat with an AI agent, Windows Hub is the shortest route. If you’re deploying to a server long-term, or wiring this into a CI pipeline where you want script control with no GUI, the native CLI is the better fit. If you’re already comfortable with the Linux tooling ecosystem — Linux-only MCP servers, shell scripts, that kind of thing — or want to keep reusing it as-is, the WSL2 path is still a fully valid choice.
🪟 There Are Three Install Paths Now
| Method | Status | One-line description |
|---|---|---|
| Windows Hub | Officially recommended | Native WinUI app, no admin rights needed, includes tray icon, one-time setup wizard, and chat window |
| Native CLI + Gateway | Valid alternative | PowerShell install script, genuinely native, no WSL2 required |
| WSL2 Gateway | Still works, the old method | Manual install, what the official docs call “the most Linux-compatible runtime” |
Worth noting: these three are specifically about what to pick on the Windows desktop. The official /install docs also list package-manager-level install methods separately — npm, pnpm, bun, curl install scripts, Docker, Podman, Nix, Ansible — but those are a different axis, a detail you choose “within” whichever of the three paths above you land on.
✅ Installing via Windows Hub (the recommended path)
Grab the signed x64/ARM64 installer from github.com/openclaw/openclaw-windows-node/releases/latest. You’ll need Windows 10 20H2 or later, or Windows 11, and it installs without admin rights. Once installed, you get a tray status icon, a first-run wizard, a native chat window, the Command Center diagnostics tool, node mode, and local MCP server mode.
💡 A precise distinction worth making: Windows Hub’s default “set up locally” flow still automatically creates a dedicated, app-owned WSL2 distro (OpenClawGateway) under the hood. In other words, the WSL2 technology itself hasn’t gone anywhere — the Hub has simply automated the WSL2 install, Node install, and npm install steps a user used to do by hand.
The same precision applies to MXC sandboxing. MXC (Microsoft Execution Containers), introduced as a security feature at the Microsoft Build 2026 keynote, is not bundled by default with the Hub or the native CLI. It requires installing a separate opt-in plugin, @openclaw/mxc-sandbox — and the underlying Windows OS container capability that feature depends on may not even be fully released as of when we’re writing this (there are reports it’s still in public preview, with a full release mentioned around October 2026 — this is information as of our check date, and may change with Windows’s own update schedule). We’re calling this out separately because we don’t want to leave the impression that installing the Hub gets you automatic sandboxing.
🐧 Installing Directly via WSL2 (the old method, corrections included)
For anyone who wants to set this up manually, or for whom Linux compatibility is the top priority, we re-verified the old post’s steps and corrected what’s changed.
- Install WSL2.
wsl --installstill works, but now there’s an added step: explicitly enabling systemd in/etc/wsl.conf. The example distro has also shifted toUbuntu-24.04. - Install Node.js. The old guide just said “Node 22.” Now you need one of
22.22.3+,24.15+, or25.9+, with Node 24 recommended. Installing via nvm is unchanged. - Install OpenClaw.
npm install -g openclaw@latest— unchanged, still accurate as-is. - Register the daemon.
openclaw onboard --install-daemon— also unchanged, still accurate as-is. - Create the Discord bot. This is where most of the changes are — covered in its own section right below.
- Write the config file. The old approach put a plaintext key string directly into the
apiKeyfield. Now the default is referencing a SecretRef object shaped like{source, provider, id}. - Verify the gateway.
openclaw gateway status/restart/doctorare all still in use as-is (anddoctornow runs more diagnostic checks than before). The oldgateway logs --followappears to have moved to a top-levelopenclaw logscommand. We’re not stating this as settled fact, though — it’s not clearly confirmed in any changelog we could find. Both forms may work depending on your version, so if the old command doesn’t work, try the new one. - Workspace files. The four files the old post covered — SOUL.md/IDENTITY.md/USER.md/AGENTS.md — are still core. But the current official docs describe a wider bootstrap file set that also includes TOOLS.md, HEARTBEAT.md, BOOT.md, BOOTSTRAP.md, MEMORY.md, memory/, skills/, and canvas/. You don’t need to memorize all of them — just check the official docs when you need to.
🔌 Discord Integration — What’s Changed
openclaw pairing approve discord <CODE> command (the code expires after 1 hour). The old post never mentioned this step at all — if you follow the old guide as-is, you’ll invite the bot, get no response, and have no idea why.
Beyond that, enabling the Server Members Intent is now recommended, and you need both the bot and applications.commands OAuth2 scopes (the old flow may have gotten by with just one). The permission set is broader too, plus you now need to turn on Developer Mode to collect server/channel IDs, and set a DM privacy toggle for pairing to work.
💡 Good to Know
Gateway port — 18789 hasn’t changed as of when we’re writing this.
Security when installing ClawHub skills — the old post leaned mostly on a general warning. Real scanning infrastructure exists now — VirusTotal, ClawScan, NVIDIA Skill Cards, SkillSpector. Rather than blanket caution, we’d recommend checking the scan summary before you install.
🧭 Wrapping Up
To sum up: there are three ways to run OpenClaw on Windows now, and for most people, Windows Hub is the simplest place to start. It’s worth knowing, though, that the Hub still sets up WSL2 automatically under the hood — the technology didn’t disappear, it just needs less hand-holding. If you’re installing directly via WSL2, follow the corrected steps above. And whatever path you take, don’t skip the Discord pairing approval step — that’s the one that matters most.