In this work multiple technologies are combined to create a complete framework to run automated or interactive user modelling experiments. Java controlled agents - optionally based on BDI platforms such as 3APL or Jason - can be introduced into a 3D game world generated by the Unreal engine. Once there, they can observe and interact with both the environment and each other. Humans also can join the simulation at different levels of interactivity. They can merely observe, partake to create an element of unpredictability, or eventually even become the main subject of scrutiny for the modelling algorithms. All behavior in the simulation is recorded for detailed (offline) analysis. By comparing the output of different predictions methods to the actual goals of the agents at the time of recording, researchers are able to quickly try new ideas, both for modelling behavior and for analyzing it.