Mixed initiative systems rely on knowledge both to know when to take the initiative and to be helpful when they have the initiative. Knowledge is usefully applied data and there is a huge amount of data available in computing systems on almost any topic. However, data is not always able to be usefully applied to a particular situation. This can occur because data isn't relevant, access to it is restricted in some way, it is not machine readable, or we simply cannot find it. We have created an architecture for making more data and knowledge available by connecting disparate user models and sources of data. We provide a comparison of this architecture to other approaches and discuss their relative strengths and weaknesses. We then describe how such systems can be used to provide more knowledge for initiating mixed initiative interactions.