Constraint programming is a modern programming paradigm devoted to solve constraint-based problems, in particular combinatorial problems. In this paradigm, the efficiency on the solving process is the key, which generally depends on the selection of suitable search strategies. However, determining a good search strategy is quite difficult, as its effects on the solving process are hard to predict. A novel solution to handle this concern is called autonomous search, which is a special feature allowing an automatic reconfiguration of the solving process when a poor performance is detected. In this paper, we present an extensible architecture for performing autonomous search in a constraint programming context. The idea is to carry out an "on the fly" replacement of bad-performing strategies by more promising ones. We report encouraging results where the use of autonomous search in the resolution outperforms the use of individual strategies. © 2011 Academic Journals.