This article is part of a special issue on Intelligent Agents in Healthcare. The authors describe a canonical architecture for an autonomous agent that can single- and multiagent healthcare applications can use. The architecture comprises a set of components whose properties are summarized as a set of software signatures specifying the invariant pre- and post-condition (input-output) pattern of each component. Components interact through shared data structures, which are defined by means of complex data types that specify standard interfaces. This allows, for example, a component with a particular design to be replaced by another with a different design. A first implementation of the model is being validated in a standard multiagent healthcare scenario developed as part of the European Union's ASPIC (Argumentation Services Platform with Integrated Components) project.