This paper describes the architecture of cooperative visual environments (CVE). This proposal stems from the findings of several experiments which suggested overcoming the limitations of first-generation user-interface management systems (UIMS) by allowing the users to determine their own computational environment. To avoid user disorientation, as well as the possibility of creating ambiguous or contradictory systems, a novel discipline for the specification and use of the tools is adopted. A systemic approach has been proposed to identify the variables needed to use, observe and adapt a CVE. The design and implementation of tools satisfying this discipline led to the definition of network objects, generalizing composite objects, and to the introduction of typed links allowing a new technique for message passing. The paper illustrates the above points by discussing the rationale behind the design of CVEs, deriving the requirements which CVEs have to satisfy and outlining the architecture with the fundamental mechanisms which allow their use and evolution. The nature of the proposal is also clarified through an example drawn from a real case. © 1993 Academic Press. All rights reserved.