So I don't know what happened, but upgrading my Docker setup to 1.0 has completely broken OH 😩 trying to create a new conversation, it spins up a new Docker container for the agent, and then becomes

Disconnected

. Browser shows websocket errors in the console when that happens. Logs for the OH container don't seem to show anything out of the ordinary, logs for the agent container mention something about

Conversation webhook post attempt X failed

, and even a fresh install doesn't seem to help. Am I doing something wrong, or should I log a ticket?

You might try switching to the CLI for a moment. Its using same v1 SDK so likely to see same issues but error messages can be easier to read from it.

Definitely file a ticket if you're unable to resolve!

Sorry I put this aside for a few days 😅 haven't tried the CLI yet but I suspect it would work, I have a feeling it might be related to accessing the Docker container from an external machine (i.e. not

localhost

). I'll try and run a test today to confirm. The following tickets seem like they might be related possibly?: • https://github.com/OpenHands/OpenHands/issues/12072 • https://github.com/OpenHands/OpenHands/issues/12083

I was holding off debugging because I knew from CLI testing I needed a newer SDK for gemini-2.5-pro... but I decided to debug a bit anyways. I can confirm both those issue you linked to were needed for me to get to point to then see LLM issue I was expected. After upgrading to v1, I can now only start agents using a web browser running on same machine as container (12072) and also only after modifying the code and rebuilding the container to add extra_hosts as mentioned in issue (12083):

extra_hosts={
                 "host.docker.internal": "host-gateway"
            },