Gave the local (but non-dev) install a spin this afternoon, so far so good. Here's a brief summary of my experience. - Before I could use it, I ran into the "Connection reset by peer" issue a couple of times initially with "Waiting for runtime". Otherwise the basic config was pretty straightforward. - Wrapping my head around the sandbox architecture took me the longest while first poking around. If there's one thing I'd suggest expanding on a bit in the docs, it's that. Basically, answering questions like: It's doing stuff without asking, is that safe? Where are my files stored? How do I get the dev tools I need? - My first test project was to try to build an app that was itself to be dockerized, so the agent immediately ran into not being able to run Docker. (It happily decided that I don't get to have what I can't have, and proceeded without. ;-) Now that I understand the architecture a bit better, that makes sense: I'd probably need a custom sandbox build for that kind of project. - I'm seeing "Agent is rate limited" increasingly frequently while asking it to do basic things. So far I just tell it manually "Continue" whenever that happens. 😛 This is with a Claude 4.0 Sonnet API key. Looks like https://github.com/All-Hands-AI/OpenHands/pull/9099 will at least make it easier to keep poking it. I've not tried hand-tuning the rate limit settings yet. That said, the sandbox is neat. I like how I can see what it's doing in the terminal, how it's taking screenshots, etc. I look forward to poking more :)

Thanks so much, this is great feedback!

Good points. We do get quite a number of questions around where are files, and how to get them.

> I'd probably need a custom sandbox build for that kind of project you mean dind or docker outside docker?

OpenHands cloud actually supports docker-in-docker! Just tell it to setup docker!

I'm on it! Xingyao Wang can track my progress here.