LangSmith Equivalent / Integration?

[AndrewKaninchen]

LangSmith has a bunch of very interesting features that I find would be tremendously useful in the context of something like guidance. In particular, the ease of dataset creation from previous runs is something I find myself wanting quite a lot. Something like that could very easily be allied with guidance for generating training data for better/smaller models that can substitute the bigger ones in production.

Are there/could there be any plans on having a native integration with LangSmith in guidance? Or, I suppose, creating an equivalent platform?

I haven't looked too deeply into what it would take to do so myself (I suppose all you'd need is to annotate guidance methods with LangSmith's traceable decorator? Maybe you'd need to do a little more, actually, to catch all the relevant metadata)

Ideas?