Another issue is If I start a new conversation and...
# general
c
Another issue is If I start a new conversation and before I type anything I use the switcher to select the LLM model I want...It will start with the default even though I asked it to start with my selection.
What logs can I provide to you guys to help? Maybe it's the docker install but the linux container is a fine resource wise. It can run for a long time working on code then it just goes to that screen after it tells me it lost connection to the backend. It's random. A screen refresh takes me back and it's still working on code so it doesn't stop that.
m
Hey @clean-postman-32056 Sorry for the late answer Could you provide the trajectory of the conversation wich surfaces the problem? You can also send me the convo privately if you want (or if the convo is private) I will look at it tomorrow as first thing i do ;)
c
I was trying to take a wiki project and rebuild it's code structure..so just normal stuff.
Also, to when this happens I have to refresh and then you'll notice the llm i chose in the chatbox and the one shown in the left panel are different. So, i have no way of knowing which llm I'm using unless i go digging through a bunch of json files. We need a better way to get this info.
I started this convo with opencode
So, it did it again. I had opencode as the llm and the backend stop responding so i refreshed it and told it to continue and it switched to deep seek even though opencode was selected.
m
Hey @clean-postman-32056 dug into this one. It's a real, previously-known bug: picking a model before your first message could get silently dropped, and the conversation would start on the old default instead. Root cause was fixed upstream, and the whole model-switching flow got reworked again just last week, so this may already be resolved depending on your build. Two things that'd help me confirm for your setup: 1. What version/image tag of OpenHands are you running (or roughly when did you last pull the Docker image)? 2. Could you share that conversation trace (privately's fine), plus — if you can grab it — the browser network tab around the moment it says it lost connection to the backend? Looking for any call to
switch_acp_model
and its response code. That'll tell me whether you're hitting the old bug (needs an upgrade) or something new in the latest build (needs a fresh look). Or you can just try to upgrade everything to be sure you are on the last version
c
The version was 1.8.0 ..the one released a few days ago. It does it at the beginning of the convo, and sometimes if I have to refresh because the backend lost connection. You can see that issue above. OH will be doing it's thing and them boom I get lost connection message with the manage backend screen. The server is fine, no resource issues and a simple refresh fixes it.
But, what happened to me was it did this and I refreshed and didn't realize the llm showing in the chat box and the one in the left hand pane were now different and it used up all my tokens with another llm that I wouldn't have used for what I was doing.
We need a debug section in the settings so we can see what's going on and get logs easily. And, in the debug perhaps a section that shows the current llm the traffic is being routed to.
A side note, I do miss the plan/code selector from opencode. Several times I've asked it to no code until told to do so ...and it just does it anyways.