This proposal describes a multi agent architecture for constructing an Intelligent Desktop Assistant (IDA). An IDA is a software tool that aids knowledge workers in performing a variety of tasks including information retrieval from multiple heterogeneous data sources, analysis of that data, construction of reports, communication with human or software agents, and a variety of other task automation activities. The proposed architecture features three communicating, cooperating agents dedicated to communications, domain processing, and intelligent user interface functions. The agent architecture is based on a proven agent programming language with extensions for agent learning and planning. The architecture features a user model that is a knowledge source describing the mental state of the user as he/she uses the IDA. The proposed IDA is well suited to managing information access over the Internet and provides the basis for electronic commerce The IDA technology developed as part of the proposed effort will form the core of a new commercial product called IntellAgent.
Keywords: MACHINE LEARNING INTELLIGENT AGENTS KNOWLEDGE-BASED SYSTEMS MACHINE PLANNING USER MODELS