Writing software for sensor networks requires full understanding of the physical phenomenon under observation. Nevertheless, in many cases knowing the details of the area of operation is difficult or unfeasible. This leads to the necessity of configuring and tuning the application executed on the sensor nodes even after the network has been deployed. We present a solution where a layer of computation is inserted between the application and the sensing equipment. This layer acts like a tiny interpreter and can be used to customize the behavior of already running applications. Experimental validation of the architecture and examples of use are also shown.