Something is somewhere. Anywhere.
The screenshot is over there, this is a more abstract description. No; this shall become a description.
From the user point of view, we see data flow through the information space and this data is processed in some form or another.This won't change until dawn. For sake of 3rd party tool support and data interchange, we technically optimize for XML data these days. But who cares? At least not the users.
"Real" users don't even care whether that processing is done by their own computer, other computers, companies or human beings. They do, however, want to keep control over whats being done with their information.
The big circle symbolizes the "information space", a network of user data and user applications. Eric S. Raymond calls this information space the "noosphere" and illustrates it in his essay "Homesteading in the noosphere". (His focus is on the rules of open source projects, which is a special case of synergy drawn from exchanging ideas, where the information space here is not centered around any pre-set concern.)
The entities of that network are called places. Places subsume what conventional call files, folders and applications. Applications could be a printers, wallets, diarys or as in the picture an user entry point, the users preferences, address data base and mail filter.
Some meta data maintained for each place. Among it slots for access control (arbitrary access control schemes - with and without a super user concept - are reasonable at the granularity of words) and a so called action, which is essentially the code of the object (from an OO point of view: data junk of the place + meta data + action = object). For more details see explanations of the grove model from SGML and DSSSL.
Two basic types of messages can be send from place to place read and write type messages. The places perform appropriate operations when they receive such a message. The read operation delivers the data at the place, possibly transformed first. It won't change the place without further help. Write operations at the other hand might change all data (and meta data) slots at the place, it can furthermore send out message to other objects and construct a reply.
Messages can be send out from a place all places can address, it might even guess. Places are addressed by a unique id or by name within a name space. Like any idea in our brain, each place has it's own name space under it's own control.
If there's no direct link (no entry in the name space for the object), but a path know via another object, the message can be send along that path. The path is expressed as an XPath.
See also the approach taken by the oxygen project, which has further details. I could not put better.