@ablaom, after the implementation of logmodel, looks like there's no plan for the mflow team to expand the artifact storage functionality (making it official).
The previous mentioned function needs the models being transformed into the MLmodel format. There's more information here. If there's any plan to make use of it, MLJ.jl needs to implement a way to transform the machines (or models) into this format.

Are you saying the previous artifact functionality has changed in a way that breaks the current MLJFlow.jl ? Or just that this API may not be supported in a future release of mlflow?

I would support work to adopt the new logmodel way of doing things. I can review parts of any work, as it applies to the MLJ API, but have pretty limited bandwidth.

Thanks for you work on this update, by the way!