Customized Content in trace when Integrating with Langchain

[lilytttt]

Describe the feature or potential improvement

When we use the callback with Langchain, all the messages content and artifact are wrote into trace. Can we customize the content wrrite into trace, i.e. can we ignore artifact or use media to replace the artifact?

Additional information

No response

[marcklingen]

You could customize the input/output by adding a data mask: https://langfuse.com/docs/tracing-features/masking

[lilytttt]

I tried the method, but even before masking the data, I already got error
WARNING:langfuse:Item exceeds size limit (size: 1293046), dropping input / output / metadata of item until it fits. and get error message inside the masking function. Any idea on how to solve this? Thx.

Try to be clear on this, we want to avoid reaching the size limit if we are working with large dataset, which is not essentially saved in the trace.

[marcklingen]

thanks for the context! let's track interest for more customization in this thread. open for contributions on this on the js and python sdk repo