A critical aspect for pervasive computing is the possibility to discover and use process knowledge at run time depending on the specific context. This can be achieved by using an underlying service-based application and exploiting its features in terms of dynamic service discovery, selection, and composition. Pervasive process fragments represent a service-based tool that allows to model incomplete and contextual knowledge. We provide a solution to automatically compose such fragments into complete processes, according to a specific context and specific goals. We compute the solution by encoding process knowledge, domain knowledge and goals into an AI planning problem. We evaluate our approach on different scenarios stress testing the main characteristics of pervasive process fragments.