Repository navigation
Replies: 1 comment
|
Note 🤖 Composer responding on behalf of wSedlacek Prior effort: #31821, #32261, #32303. What we learned trying it
|
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Summary
This discussion asks for a direction decision. Should a thread have a first-class Insights view — timing of turns and waits, tokens/estimated cost by turn, and context occupancy over time — so you can debug a slow or expensive thread without leaving it?
This is complementary to global Usage (#16722). Prior slices: #31821, #32261, #32303.
Why thread Insights
Global Usage answers “which provider spent money.” It does not answer “why was this thread slow” or “which turn blew the context.”
Agents run for hours. Operators babysitting a runaway turn need a per-thread timeline: active work vs waiting vs tooling, cost stacked on the same axis, and how full the context window got. That belongs next to the conversation, not only in a settings Usage page.
Proposal
Reference implementation
#31821 (timing flame chart), #32261 (tokens/cost by turn), and #32303 (context occupancy) are prior attempts. Linked as evidence only; prefer sliced follow-ups over one umbrella.
Related: #16722 (Usage broken down by project/thread/account) for the global side of the same question.
Question for maintainers
Yes, no, or not now: a first-class per-thread Insights panel (timing → cost → occupancy), separate from global Usage.
All reactions