Abstract

Computer networks like the Internet provide a novel infrastructure bearing new possibilities for the realisation of service platforms, while raising new requirements as well. Suppliers, brokers, and consumers of goods and services profit from their flexibility and dynamics as well as from the independency of time and location. To benefit from this potential, the open, distributed, and heterogeneous character of the networks needs to be utilised by new technologies. A promising approach are software agents, which are able to act and interact as autonomous specialists in the networks. CASA is an architecture for multi-agent systems that offers a modular structure and agents with a high degree of flexibility and interoperability. Thus, it provides a suitable basis for the realisation of service platforms in computer networks. CASA is divided into three design levels for agent composition, agent control, and agent societies, which are completed by the formal language CAL for agent specification. Individual agents are assembled from reusable components depending upon their tasks. Even at run-time, components can be reconfigured as well as added, removed, or exchanged without affecting the current work of the agent. Thus, agents are created and adapted to changing tasks easily. Agent control empoys a knowledge-based control scheme that integrates reactive, deliberative, and interactive behaviour. Formal knowledge structures provide an abstract layer for the specification and analysis of agents and form the basis of a flexible, autonomous behaviour control. In addition, formal knowledge is utilised as the content of communications and to describe the interactive capabilities of agents, thereby enabling the interoperability required by agent societies. Interactions are represented as service acts provided and used by agents in order to delegate tasks. In combination with an agent infrastructure that administers agents and services, the service scheme allows flexible and dynamic collaborations among agents and the integration of interactions into the behaviour control. The agent description language CAL is a combination of formal logic and programming language concepts. It covers all aspects of the representation, use, and communication of knowledge by agents. CASA is an open and scalable architecture to build flexible, dynamic multi-agent systems that supports modularity and interoperability at all architectural layers.