OpenHands

I think one blocker for agent canvas usage is testing on Windows. I’m getting several people interested in windows but I think our support is not great. cc <@U0AM1NT36R2> so we can think abut planning around this

I think this shouldn’t need be manual, but I don’t know a source of Windows machines, maybe GitHub?

That’d be awesome, e2e tests, integration tests, or evals using linux and windows that show the gaps would be great.

I have my gaming/personal windows machine where I've tested and run the electron app. I also have a few old windows 10? I think ones I can manually test on. But automation would be preferred for sure

Yeah, I have a windows machine I should use more

Ideally, a Windows cloud machine would be nice, I just dont know a source

GCP has google cloud virtual machines, we could probably spin one up

I do have an issue assigned to me for that

And gcloud don’t give virtual machine really as we would like to test it

One things is going with gh actions for example

I suspect gh actions is not enough for my fleet, actually :thinking_face:

GCP machines allow you to test a windows desktop, but they wouldn’t be good for CI

<https://github.com/OpenHands/OpenHands/pull/19246|github.com/OpenHands/OpenHands/pull/19246>

&gt; GCP machines allow you to test a windows desktop
Yes yes. For instance we can reproduce in the CI all the bugs reported. I mean we could eventually set up an automation which go into the machine and update to the main version and do QA. This could be something

and you run just before cutting a release

Yes, I would try to have agents run openhands on Windows, and test all it can do, like a user would.

The bugs they find, they report, fix, add regression tests in CI