Loading a large cloud convo took ~2 min I wonder i...
# proj-agent-canvas
q
Loading a large cloud convo took ~2 min I wonder if we can do smth to speed this up
c
Oh, thank you for this video. The same is happening on current Cloud platform, and I meant to capture it, because this doesn’t feel like the right experience
Alternatives I’d think of: • load only the last LLM context (so from the last condensation) ◦ maybe paginate the rest • paginate as normal, as in, e.g. every 50 events • ask for and load the last ~20 ish events, and get a summary for the previous stuff ◦ but this… is kinda already made for us, by point one
IMHO we’re moving away from the times when OH needed to show the user every tool call You know, with long running conversations, people are not going to be even able to look into every tool call in history, and if they do, they probably need an observability platform or other ways to manipulate the events files rather than a Web UI for everything
nod 1
Another alternative that has been suggested at some point has been to display the plan there, the old plan. Personally I’m kinda in doubt about that these days 🤷, because it’s not always as linear as it sounds, and in a long running conversation there may have been more than one. Idk maybe… but well, that’s… kinda similar role that a summary performs?
nod 1
w
I had lazy loading working really well for local convos! It's currently broken though, it scrolls all the way back to the start of the convo forcing all events to load
I think it was working well for cloud convos too but I can't be sure
c
Once saas is released, we can merge this which should help https://github.com/OpenHands/agent-canvas/pull/407
🙏 2
previously event pagination is broken due to one silly bug 😅
ok actually looks like cloud release is shipped yesterday, will merge this to main now!