A jest-in-time information retrieval agent (JITIR agent) is software that p
roactively retrieves and presents information based on a person's local con
text in an easily accessible yet nonintrusive manner. This paper describes
three implemented JITIR agents: the Remembrance Agent, Margin Notes, and Ji
mminy. Theory and design lessons learned from these implementations are pre
sented, drawing from behavioral psychology, information retrieval, and inte
rface design. They are followed by evaluations and experimental results. Th
e key lesson is that users of JITIR agents are not merely more efficient at
retrieving information, but actually retrieve and use more information tha
n they would with traditional search engines.