I work at QA Wolf and lead the development of automated testing for Windows desktop apps. I wanted to share what we did and how we did it.

This post covers how we added Windows to a test infrastructure built around Linux runners. The central problem was giving each test a clean, logged-in Windows desktop without waiting 2–3 minutes for a VM to boot.

We extended the device scheduler we already use for mobile testing. It maintains warm pools of Windows VMs on GCE. Each test claims a ready VM, uses it once, and releases it. The VM is then destroyed and replaced, rather than cleaned up for the next test.

The test code stays in a Linux pod. Playwright connects to browsers on the Windows machine over CDP, while WebdriverIO and Appium handle native apps through WinAppDriver. A small Go agent captures the desktop with FFmpeg and streams it over WebRTC for live viewing and recordings.

That separation lets a single test move between a browser and a native Windows app without requiring a separate Windows runner pipeline.

The trade-off is that we keep idle Windows VMs ready and install the app under test on every run. In exchange, tests don’t inherit another run’s files, registry changes, or saved sessions.

I’d love to answer any questions!