Hey,
(Ref: <https://github.com/OpenHands/agent-ca...
# proj-agent-canvasm
modern-fireman-99755
05/18/2026, 9:14 AMHey,
(Ref: https://github.com/OpenHands/agent-canvas/issues/570)
I was playing around with agent-canvas and noticed that tool calls aren't displayed very well at the moment (see the video in the issue).
I think presenting tool calls in a comprehensible way to the end-user not only enhances the user experience, but also makes it easier to debug, perform QA, and create good snapshot tests. For these reasons, I did a bit of scoping on how we could enable that. You can find all the details in the issue.
Looking for feedback on the design, implementation, or just general thoughts!
c
calm-artist-7206
05/18/2026, 10:47 AMNice! I love the custom visualizers!
You may laugh, but I actually come to this from the other side: I doubt people will want to look at the exact output of a tool call a lot. In fact, a nice UX might… not be the DX of a debugging tool. Hey, we have debugging or observability tools for that 🤔
I understand we may want to see details now, because we’re just building stuff; and it seems for ‘task’ and other tools that were new, we don’t have some nice shape. IMHO though, that’s different than both
• what is nice for the user to see in this conversation - I… think it’s more of a conversation, than a debug session
• what is the… how do I put this, what is the likely direction in the next ~six months for how people work with / expect from agents -> I feel strongly that it’s more for long running tasks, where you can’t keep up anyway… you need real observability, when you need it; or nice high-level view, when you don’t…
calm-artist-7206
05/18/2026, 10:48 AMBut that’s why I love the direction you took: give it to people. Give people the ability to make and add the visualiser they feel like
calm-artist-7206
05/18/2026, 10:51 AMI could see an argument for going the other way, too:
https://openhands-ai.slack.com/archives/C06QT0AGY4W/p1778946098902129
calm-artist-7206
05/18/2026, 11:05 AMI mean, maybe we can split like:
• a default visualizer is necessary
◦ we could keep it simple
• give people the right tools, to create for themselves a visualizer
◦ cc @hundreds-twilight-32833 ‘s addons maybe
◦ without code modification
• but even so, the events become too many
◦ easy to see in restoring long -ish conversations scenario
• so whip up lazy loading of the full events payloads like @wooden-umbrella-84098 did
◦ then, the visualizers for the older events would show them like the latest events
• or maybe go even further, and have a summary of “what went on before”
◦ presumably in conversations view
◦ or replace planner view with it (in some conditions?)
◦ in human, not in tool calling output, and not in full terminal output 🙏
👀 1
h
hundreds-twilight-32833
05/18/2026, 12:33 PMI've been thinking about a Timeline view for conversations. I wonder if we could reduce the agent tool calls and activity down to a single line of 'Thinking -> Working -> Worked for' and when you click that it opens a right hand panel with a more detailed and visual timeline of what the agent is doing. Maybe that would also improve the conversation load time and overall performance.
w
wooden-umbrella-84098
05/18/2026, 12:59 PMI would love to see thoughts explicitly displayed! I think we're just dropping them from the UI now?
Here's what I have in my custom UI
m
modern-fireman-99755
05/18/2026, 1:04 PMwe do have thinking already displayed as a tool call
👍 1
h
hundreds-twilight-32833
05/18/2026, 2:00 PMCurrent Thinking for reference
2 Views