Hey, PR to fix the problem with deleting the last...
# proj-agent-canvas
m
Hey, PR to fix the problem with deleting the last profile: https://github.com/OpenHands/agent-canvas/pull/1128 Moreover, if you don't have an LLM in our store, you can't start a conversation, or rather, you can prepare it, but you won't be able to send messages. This prevents unexpected behavior. @hundreds-twilight-32833 Can you try to break it?
🫡 1
h
Any specific testing instructions?
m
not really 🙂 just delete all the profiles and try to start a convo
h
Only thing weird I've found so far is trying to change a profile during a long conversation. But I don't know if that was a previous issue or not.
m
what do you mean? video?
h
Recreating it now
well now I can't reproduce it ¯\_(ツ)_/¯
🎉 1
Minor renaming issue
m
Ah ok I see... but this is unrelated to the PR. It is just on how we are validating the profile names. We are probably normalizing them.... We can do that in another PR 😉
h
Sounds good! I'll keep poking at it for a few more minutes
m
Thanks
h
LLM profiles are important. I definitely want to help make them a great experience 😄
m
Hey @hundreds-twilight-32833 How are the tests doing? @quick-window-91705 it seems the e2e tests are failing but are not related to my changes (automation and skills). What should I do?
q
skills can be ignored, comes from some other pr that merged without a fix for automation its bc we setup an llm profile prior to running the automation, and the behavior of profiles has changed a bit causing the test to fail have oh take a look
👍 1
h
I have not been able to cause any new errors. You mentioned trying to delete all profiles but I am unable to delete my last profile because it is active which is a good thing I think
m
@hundreds-twilight-32833 are you sure that you are using the last version of agent-server?
1.25
h
Oh not sure now. I can pull back down
I do see it now. Retesting
q
Do we still pass full settings objects when creating conversations? Or do we pass just the active profile name?
m
why? Yes
h
Nothing on either of my machines came up @modern-fireman-99755
m
it means is good right?
🙂
q
just confirming the expected behavior for automation, it fetches the current agent settings at runtime. but if we activate a profile and don't immediately patch the settings it may not get the active profile?
h
Yes, no issues found when messing with LLM profiles. I didn't check automations however...
q
^ also i'm not completely sure if thats the case, just hypothesizing a buggy scenario
we may need to start specifying the profile the automation should run with, as the user can change the active profile frequently @hundreds-twilight-32833 may want that to be in mvp
h
So does it use whatever the active LLM is right now?
nod 1
q
Confirming, but yes I think so
h
Something to change for sure. Ideally you can set an automation with a cheap LLM and it always uses that cheap LLM. I'd like to figure out some new pattern than just saying everything is MVP. I'll create a ticket
👍 1
q
^ confirmed we always use the existing settings the user can't say "always run this automation with gpt" if they switch from gpt to claude suddenly their automations switch as well (this is both for local and cloud automations)