We consider the problem of enabling read-only access to a very large number of independent databases over a public network without compromising the autonomy or heterogeneity of the participants, both databases and queriers, who may number in the thousands or millions. From an examination of the network environment and its characteristics we derive general principles for designing applications involving autonomous systems, including statelessness, decentralization of control, and simple communication standards. Applying these principles, an architecture is developed for information exchange among diverse database systems. This architecture preserves autonomy by separating the functions of query interpretation, database discovery, and user interface into modules independently executable by any participant or subgroup of participants at any time. Subproblems arising in these three areas are discussed in detail, and solutions are presented.