@wooden-umbrella-84098 / @limited-student-10747 - Do you have any objection to having the `README`a show docker instance being completely separate from npm? • Using a different home directory (So that the openhands config / conversations are separate) • Using a different port (So that browser local storage is separate) The reason I ask is that there are a whole class of bugs that come from swapping between docker and npm due to absolute paths to resources within conversations, and this is a great way to cut scope.

docker run -it --rm \
  -p 8001:8000 \
  -v ~/.openhands-docker:/home/openhands/.openhands \
  -v ${PROJECTS_PATH}:/projects \
  <http://ghcr.io/openhands/agent-canvas:<WHATEVER|ghcr.io/openhands/agent-canvas:<WHATEVER> IMAGE YOU'RE TESTING>

I understand, though… to be honest, I don’t love this idea, because a significant amount of user pain out there has been due to different configs and unrecognized conversations/settings for what was supposed to be the same OpenHands: between CLI and Web, between running from this or from that… 🥹 From what I understand, this is exactly what this would do, e.g.: • run with docker, set your stuff • run without, can’t find your stuff • modify one -> not recognized in the other 🤔 I don’t know if there is a way to make it understandable…?

I think in the long run it would be nice to have these common, but there are issues which prevent it - primarily absolute file paths in the conversation files - so a conversation created outside docker cannot be loaded inside and vice versa. I don't see to many end users jumping between docker and npm - though I could be wrong

I think people definitely jump, since they jumped between CLI and Web and headless - and Cloud, Cloud is still a bit strange since we lost the ‘user skills’ repo. Thinking maybe • conversations could be in different dirs…🤔 • but the same settings are necessary Why would settings not be the same? This one has been very surprising, in a bad way, to people

One alternative I have from Saturday is https://github.com/OpenHands/agent-canvas/pull/960 In this case only the conversations are not shared. I figured this was more risky than full separation - particularly when people have different versions of openhands in their repositories (How would one downgrade the state between npm and docker?) But maybe it is worth the risk

Approved, thank you!

e.g.: • I have a version of agent-canvas locally synced from main - it is using a newer version of the agent server. • I use a docker command including an older version of the agent canvas (with an older agent server) • I start a conversation in the npm from main. • I then stop my server, and start the docker one. I have no idea what will happen 🤷

I think maybe we can come back at some point to the idea of ‘teleporting’ conversations, I haven't seen it recently around but it might still be cool- and solve these ‘missing conversations’

Yeah for now, I'm good w/ splitting the docker and npm configs. I don't think people will be switching back and forth, and there's a reasonable migration strategy

Just for clarity, the split is for conversations only

oh even better