This thesis proposes the cognitive ensemble reference architecture. Cognitive Systems attempt to map human thought processes onto technical information systems. Achieving this for coordinated device ensembles is hereby considered a particular challenge. Based on a specification of (software) components and a coordination metaphor utilizing live, semantic annotations, this reference architecture represents a template for the implementation of cognitive behaviors in ensembles comprised of heterogeneous, networked computing devices. Previous efforts to bring together and take forward these fields of research have either been too specific in their implementation thereby failing to achieve wider application, or been hindered by user acceptance, lack of interoperability, and complexity of participation. This thesis establishes a very basic definition of what cognitive ensembles are and how they may be created, based firmly on accepted knowledge from the fields of artificial intelligence and cognitive psychology. Multiple exemplary ensemble instantiations are described, used for investigation of research questions and evaluated according to the outlined aspects and properties of the cognitive ensemble reference architecture. This thesis constitutes a contribution to artificial intelligence research and cognitive psychology, specifically in the domain of cognitive systems, live tuple spaces, wide-scale adoption of IoT-style systems, multi-agent coordination and design of reference architectures.