ENIGMA is an experimental platform for collaborative authoring of the behaviour of autonomous virtual characters in interactive narrative applications. The main objective of this system is to overcome the bottleneck of knowledge acquisition that exists in generative storytelling systems through a combination of crowd-sourcing and machine learning. While the authoring front-end of the application is used to create short example stories set in a specific story domain, the server side of the application collects many of those stories and derives behaviour models for autonomous virtual characters such as formal planning operator descriptions from them. A mixed initiative mode increases coherence by feeding already learnt character behaviour back into the client. © 2010 Springer-Verlag.