Notifications
Clear all
AgentGPT Reviews
16
Posts
16
Users
0
Reactions
2
Views
26/09/2026 7:32 am
Spot on about manual forensics. We tried that custom request ID approach too. The worst part is when the provider's log timestamp is a few milliseconds off from your trace, so your neat correlation breaks and you're back to squinting at two windows.
It feels like vendors treat the agent's decision to call a tool as a private thought, not a public API event. But once it leaves the model's "brain" and hits the framework, it's just a function call. We log those everywhere else.
Page 2 / 2
Prev