Policy-driven management of agent systems

The agent paradigm has the potential to help in the development of applications for the open and heterogeneous Internet environment. Agents acting on the behalf of users can autonomously fulfil assigned goals, thus relieving users from complex and time-consuming tasks. Agent-based applications typically involve multiple agents, and each agent has to play a specific role that defines what the agent can and must do in order to achieve its application goal, and how it interacts with other agents and with the environment. This paper describes the integration of a policy language (Ponder) within an agent infrastructure (SOMA) in order to flexibly model agent roles and agent behaviour according to application specific requirements. The Ponder language is used to specify both agent permissions and duties and to model agent behaviour in terms of roles and relationships. SOMA is a rich infrastructure to support agent execution and provides a set of facilities that can help to build Ponder policy enforcement services. The integration of the two provides a flexible framework to the development and management of agent applications.