botchat.jpg shows a sequence of cartoon messages between parties talking Jabber; the following describes
the interaction shown in the image (with a few details added).

bandri@livejournal.com/jqbus (shown as a laptop) asks the Jabber service, "Who is on my roster"?

Jabber service replies to bandri with a list of 5 Jabber "JIDs" (identifiers), which include
sub-resource names. So we see danbri@livejournal.com/ichat 
danbri@livejournal.com/jqbus libbymiller@gmail.com/jqbus libbymiller@gmail.com/jqbus alice@example.com/Adium and bob@example.com/tvp

We then see bandri ask "Who speaks SPARQL?" to each JID on his roster, using "DISCO" (the XMPP 
service discovery spec). The diagram shows one response, from libby@gmail.com/jbus confirming that 
it "can do http://www.w3.org/2005/09/xmpp-sparql-binding" (IQ extensions are identified by URI).

Next bandri sends a SPARQL query to that JID. Note that we skip over any form of Semantic Web service
discovery here, and any access control. In practice, this step is where we might ask for a list of 
"named graph" datasets, and determine whether they are accessible. The SPARQL shown is simple, and addressed 
as an IQ 'get' to jabber:libbymiller@gmail.com/jqbus:
 
SELECT ?name, ?nick WHERE { [ a :Person, :name ?name; :nick ?nick ] }

A tabular response send to bandri is shown, with columns for the variables ?name and ?nick, and a row 
for each Person description that matched.