This paper describes a prototype application which combines speaker identification technology and an agent architecture to provide user-definable monitors for incoming voicemail messages. Through a Web-distributable Java user interface, the user may enter requests by using spoken or typed natural language. Multiple distributed agents process the requests, periodically testing the user's voicemail system to identify the composer of the message from a set of selected speakers. When a message meets the conditions specified by the user, agents locate the requester's position and notify him or her of the arrival of the important message by using various communication media (email, fax, telephone, pager). The technology responsible for identifying a speaker from voice is a text-independent method developed at SRI International. Encapsulating this capability as an agent permits plug- and-play reusability in the growing number of applications being developed within the agent-based framework.