Prioritize traces captured from LangGraph by relevance

[litagent]

Describe the feature or potential improvement

Is there a way langfuse's CallbackHandler capture the necessary traces generated when using langgraph?
similar to langsmith screen shots attached. both langfuse & langsmith capture lot of unwanted information with null values which ideally needs to be omitted

Additional information

No response

Contribute

• Yes, I can implement this and raise a PR

[marcklingen]

Thanks for raising this idea of adding a Most relevant filter to get less verbose outputs in Langfuse. This seems to be a problem mostly related to the Langchain integration as other users of Langfuse define more deliberately what they'd like to add to a trace.

Is there a good heuristic of what's important/interesting to you in a Langgraph trace?

[phenomagentx]

LLM & tool related traces(start & end) are the most relavant for any agent implemented using Langchain or Langgraph @marcklingen

[hookenful]

Vouch for this

[marcklingen]

thanks again for adding this here, totally agree that some of the observations should rather be hidden by default.

in the meantime, @jannikmaierhoefer just contributed a notebook to the documentation with more LangGraph examples (just fyi): https://langfuse.com/docs/integrations/langchain/example-python-langgraph