The recent development of new terminals such as phones, mobile computers, e-books, etc, raises the needs for new interface modalities, in order to replace or complement the traditional mouse/keyboard interface. Ideally, these new interfaces should use limited computing resources and should be easy to adapt to a specific user and to a large variety of user needs. We propose here a new handwriting recognition system that is an attempt to handle these constraints. We evaluate its performances and ability to adapt to new users on a part of the Unipen database.