We discuss a virtual presenter in the form of a 3D avatar whose task it is to show people a way through a (3D model of a) public building. Our thesis is that when looking for way descriptions, seeing such a walkthrough guided by a virtual scout is easier to remember than a purely textual description of the way or tile kind of annotated floor plans used widely in today’s public buildings. Furthermore, we discuss some technical aspects of how such a scout can be constructed and what underlying data and processes are needed for the automated generation of guided way descriptions. One of our main goals is to give the avatar the possibility to react to her spatial environment in an appropriate matter, e.g. pointing to relevant objects and following a path that can be easily memorized by tile viewer. Tile walkthroughs described here will be used within a building navigation system that also includes navigation via a handheld device, (such as a PDA,) once tile user has left the information booth. Both the 3D visualization at the info booth and the more sketch-like presentation on the PDA are generated from the same data and by the same system, and thus can refer to each other for a much more consistent overall appearance.