OMax listens to a musical stream, automatically detects significant features such as pitch, timbre structure, duration etc., builds up a vocabulary of units based on these features, organizes the musical units into higher level formal motivic structures, and finally plays by itself by navigating the structure and generating back a stylistically coherent stream of music. This results in an on-line smart segmentation and recombination of the input sound material. All these processes happen very fast and simultaneously, so that when used live in real-time interaction with a musician, OMax gives the flavor of an immediate co-improvisation between the performer and the machine. Because the performer also reacts to the machine’s production, thus impacting on the model’s evolution, it may install a double feed-back that we call “stylistic reinjection”. OMax has been performed all around the world with such great performers as Bernard Lubat, Steve Lehman, Maria Kimura, Vijay Iyer, François Corneloup, Raphael Imbert, Lori Freedmann, Rohan de Saram, Mike Svoboda (through the HexaGram Native Alien System), Brice Martin, Mederic Collignon, Cécile Daroux, musicians of the Ensemble Intercontemporain, Denis Beuret, Mike Garson, etc. It is one of the more powerful and most used machine improvisation systems today.