This paper describes our work on a portable non-invasive brain-computer interface (BCI), called Adaptive Brain Interfaces (ABI), that analysis online the users spontaneous electroencephalogram (EEG) signals from which a neural classifier recognizes 3 different mental states. The outputs of the classifier are used as mental commands to operate communication and control devices. Although still at a research stage, BCIs offer the possibility to augment human capabilities in a natural way and are particularly relevant as an aid for paralyzed humans.