This document reports on an architecture for commonsense thinking presently under development at MIT by Marvin Minsky and Push Singh. This architecture is being implemented as a set of software components that together will be capable of reasoning about the types of'common sense'concerns that show up in most domains, for example, those that require reasoning about temporal, spatial, physical, psychological, social, and reflective matters. Implementing such a wide range of abilities requires a framework that can make use multiple strategies for representation and reasoning, for ordinary commonsense thinking spans so many different types of problems and depends on so many forms of knowledge that more unified frameworks, ones that primarily make use of a single type of representation and mode of reasoning, are stretched beyond their capacity. Just as biological systems have no single, simple principle for their operation, we expect that cognitive systems will contain just as diverse and heterogeneous a variety of components.