Please merge or improve this PR: <github.com/OpenH...
# general
b
Please merge or improve this PR: github.com/OpenHands/software-agent-sdk/pull/4028 It's super important, because it is causing timeouts and stopping inference.
c
Are you usually switching LLM during conversations?
In my understanding, this PR only applies if the agent uses switch_llm tool or the user switches profile
Am I wrong?
b
Not yet. I tried it once, and since then, many of my conversations reset the timeout to a too short value after agent-server restarts.
I had two profiles. I corrected the timeout of one profile, but after every restart, the timeout resets to a value that I don't know where it comes from
but it's not set in any LLM profile
it's like the default timeout used when the convo was created
c
What I mean is: lets say you start a new conversation now. Do stuff; then close the tab. Come back and restore it: if you never switched LLM during that conversation, I don’t think the PR has any influence
So if the timeout still resets for you - so it’s not respected as you set in profile, then the bug seems different
b
How did you setup the initial LLM setting for this convo?
If you do not touch the LLM settings, I guess the convo will have the same initial settings all the time
c
When you first start agent-canvas, a wizard asks for an LLM details. Then creates an LLM profile
Then all conversations would start with that profile, unless you switch
b
But if you want to increase the timeout, because your LLM engine is stopping the convo, you probably won't be able to do it
c
That profile has timeout, though, you can set your desired value there, you did that?
Why not?
b
I set the timeout in the profile and it works fine, until the conversation is restarted
after restart, it is reset to some initial value (that you probably cannot change anymore via the UI)
c
During that conversation, did you use this dropdown to switch to a second profile?
b
yes, when I change timeout, I switch to another dummy profile and switch back
and I confimed that base.json is changed
Copy code
cat /workspace/conversations/$SESSION/*.json | jq . | grep '\"timeout\"'
When agent-server starts, it changes the base.json again
with a unknown timeout valule
In the linked issue, I provided instructions to reproduce
super easy, just change the timeout and restart the convo
and take a look at base.json
I'm using know agent-server with only two patches
c
https://openhands-ai.slack.com/archives/C06P5NCGSFP/p1783792783106159?thread_ts=1783791409.275459&amp;channel=C06P5NCGSFP&amp;message_ts=1783792783.106159
You only did this once, right? There is no need to do ever do it again, just let the Active profile be the profile of any new conversation
(TBH I think it should never be needed, but at most once if we don’t refresh properly on profile save.)
b
You only did this once, right? There is no need to do ever do it again, just let the Active profile be the profile of any new conversation
I'm not changing LLM profiles. I only need one setting. The issue is: when I restart an existing conversation, it resets the timeout to it's initial value.
it's not using the timeout from the profile