In many parts of the Intelligence Community2 data gathers at rates that preclude the ability of analysts and policy makers to ingest, let alone comprehend, the complex information confronting them without working together in teams. These teams often form ad-hoc, mission-oriented communities of practice, and sometimes larger multi-mission communities of interest. These communities can be distributed across the globe. Within these communities analysts try to share their insights but often end up simply passing around partially processed information. This paper reports on developments toward a next generation collaboratory supporting small groups of analysts (or policy makers) engaged in joint analysis and problem solving. The work reported in this paper "automates what ought to be automated," differentiating between what machines do well and what humans do well. It describes an infrastructure that automates much of the information exchange between analysts while at the same time creating an environment that facilitates insight sharing among analysts in a community of interest. The architecture of this infrastructure is called an upstairs/downstairs collaboratory where "upstairs" is an analyst environment focused on providing access to colleagues with special resources and building understanding of alternate world-views to facilitate the sharing of mutual insights; and "downstairs" is an automated infrastructure for information sharing comprised of an innovative information store serviced by software agents. This infrastructure manages multiple contexts and rapid change by relying on multi-layered models. The models are built on a flexible storage structure that allows late semantic binding for fast processing of model changes and differences. An initial prototype has been built and is currently being extended.