What you are trying to do goes against the core concept of Dataiku: that every transformation step (ie: recipe) must have an input and an output defined so it can be displayed in the flow. You can write free style Python code and define your input and outputs dynamically but run it your code elsewhere. There are however several ways to work around Dataiku guard rails. However before suggesting any of them I think you really should clarify your actual requirement. What exactly are you trying to achieve? Why do you think you need dynamic inputs? When you say different "workflows" what exactly do you mean? Is that another Dataiku project's flow? Where are these datasets stored on? Are they of the same structure?
