Consider using lmnr.ai and their observe() method to see if I can trace input/output from method calls.

[burtonator]

Better than building this by hand PLUS I get the function call duration.

https://www.lmnr.ai/

[dinmukhamedm]

@burtonator not sure you've created this in a correct repo :)
But very curious to learn about your use case

[burtonator]

@burtonator not sure you've created this in a correct repo :) But very curious to learn about your use case

Hilarious! You're right. Wrong repo. Sorry. Was just to fall asleep and wanted to make a ticket for this!

Anyway. I wrote something internally and it's for a workflow agent (not fully autonomous agent).

I want input/output tracing and a performance UI but the other 50% of this is that I need resume so that if the same job starts up it can resume from cache and continue executing.

It would also be nice to view cache dependencies as a graph but that's fore the future.

Most of the functions I want to observe are "slow" meaning 500ms-5000ms

Happy to discuss..

[skull8888888]

@burtonator happy to help you get started with Laminar. It's actually extremely easy because we auto-instrument majority of LLM providers and frameworks. Check out our getting started guide here https://docs.lmnr.ai/tracing/introduction. If you have any questions ping me anytime