Swapping between normal flow and OpenAI assistant

Hi everyone, does anyone have an experience/suggestion on how to mix “normal” calls with OpenAI assistant (predominantly Code Interpreter). We built all our flows around normal model calls with auto function invoke, but we want to add some spacy things like creating a graph from the data.

In particular, my question is, should we wrap an OpenAI assistant call as function with a description like - if you do not have the skills to answer the user’s request, call this function (and hope this function will be auto invoked will ) or any other alternative?