Hello, I've been a lurker of this project for a while. To get some clarification on @limited-student-10747 message above: Are you guys are planning to move https://github.com/OpenHands/software-agent-sdk into https://github.com/OpenHands/OpenHands, and break apart what is currently https://github.com/OpenHands/OpenHands into the other repos you mentioned?

Yeah,

OpenHands/agent-server-gui

will basically be what’s within the

OpenHands/OpenHands

repo’s

frontend/

directory, but it will connect directly to the

OpenHands/software-agent-sdk

server directly to make it super easy to run operationally, make local deployments lighter, etc.

I want to make sure I understand the intent behind this change, because the shape of it is giving me some pause. If the goal is to separate out the proprietary logic, that seems achievable by breaking it into its own repo (or repos) without moving

OpenHands/OpenHands

at all. The proposed migration is a much heavier lift for what should be a scoping problem. Issues, PRs, releases, and tags would all need to be triaged or migrated, half of them referencing files that won't live in the renamed repo anymore. Forks lose a natural continuation of the source they were tracking. Every external reference (package manifests, docs, blog posts, Stack Overflow answers, papers, etc) keeps resolving via redirect, but to contents the original authors never linked to. That's a lot of downstream disruption to absorb, and I'm not seeing what the migration buys us that a clean extraction wouldn't. Happy to be wrong here. What am I missing?

Hey @victorious-market-46627, actually I spoke to Robert this morning, and he agrees with you so I think we might revise the plan. Thanks for the feedback!

> The proposed migration is a much heavier lift Yes, it is. I have worked on it and presented a quick list here: https://openhands-ai.slack.com/archives/C086ARSNMGA/p1776394217399069?thread_ts=1776196951.050539&channel=C086ARSNMGA&message_ts=1776394217.399069 . Please find, for example, how I forwarded issues/PRs in the Web repo. I believe it’s worth it. You are correct if it was only about some code. To my mind, though, it’s about the OpenHands project, its definition, its people, its culture. We need to refactor repos in such way that it set OpenHands up better for the future. Merging the repos (sdk with oh/oh) implies merging contributions, too. While a new repo would mean wiping the 500+ authors of OpenHands, the Open Source project? Or the SDK? We cannot and should not do that, I believe. For clarity, we had settled on merging the sdk in oh for 5+ months. It’s not easy, but it’s not new. 😅 Personally, I will honestly admit, with apologies, that I think it’s important and rely on it. It’s also tied to other proposals we were discussing, to change permissions and ways of working, so that we get, in OpenHands org: • open source process in open source repositories. Long story, sorry 😅

Hey Engel, thanks for the perspective! I think you saw my revised proposal above, but to put some more color onto it: I think the agent-canvas and software-agent-sdk repos have different audiences, where agent-canvas is more user-centric, and software-agent-sdk is more developer-centric. Up to this point, OpenHands/OpenHands has been user-centric, and software-agent-sdk has been developer centric. According to our research, many people start out using the GUI and then either continue using the GUI, or transition to using other form factors. Fewer people start from using the SDK. So the thinking behind the revised proposal is because (1) the GUI is the most common entrypoint into OpenHands, and (2) OpenHands/OpenHands has traditionally housed the GUI, it seems like that'd be a good thing to have in the representative OpenHands/OpenHands repo. Note that I didn't originally think this (I was originally in favor of having software-agent-sdk in the main repo), but hearing about Devin's discussions with users, and also some chats with Robert changed my mind. I'll let Robert chime in a bit too

I hear you, and to be honest, I was thinking too that something additional to the SDK would be nicer than the SDK alone. An end user could have an easy install, that doesn’t require the UI codebase in the repo too. Maybe like this: https://openhands-ai.slack.com/archives/C06QT0AGY4W/p1778343731980929?thread_ts=1777582173.007339&channel=C06QT0AGY4W&message_ts=1778343731.980929

Got it, I'll let others pipe in with ideas too!

I see Engel’s point. I don't have a great understanding of how all this works mechanically at the git level, but I like the idea of the main OpenHands/OpenHands repo being something that represents ~sum total of the community contributions. We want to grow contributions and devs like contributing to the main repo with all the stars. I also agree with Graham that new users of OpenHands will most likely start with the Agent Canvas GUI first, but eventually graduate to SDK once they want to go deeper. However, if SDK becomes the main OpenHands/OpenHands repo, I think we could still promote Agent Canvas GUI as a starting point in the README — and some easier install instructions other than “git clone agent-canvas” Ultimately both options have pros/cons. My vote is the path that attracts contributors and gives credit to community. Of course we want contributions to both agent canvas and SDK, too. This might need to be something the maintainers sort out or have Robert be the tie breaker here.