In this article we present an architecture for the integration of tutoring approaches and process scaffolds into existing collaborative applications. The architecture allows to combine existing research on explicit representations of collaborative learning processes with the availability of existing and tested collaborative learning environments. The architecture allows to control the learning environments either by a human or a pedagogic agent and thus enables adaptation of the tools to the current state of the learning process. Both types of tutors are using a so-called remote control component using the same set of control primitives. To prove the soundness of the architecture and the flexibility of its implementation two example scenarios are shown that use IMS LD learning process definitions with the Coppercore learning design engine controlling our collaborative environments.