I'm assuming - and this could be a bad assumption - but based on the continuing updates architecture, if client a changes model a, client b will see an update on model a. How does client b get notified of the change? Does it have some kind of fallback system ala socket.io? Is this not yet part of the project?
Judging by the chart on the site, it looks like each client polls the original model on the server, and infers changes by diffing it against the client's current copy.
It tries to be agnostic about this and favor any particular push method. Instead, it simply provides the primitives on the client to make these things easy (e.g. merging in new data).
It looks like it tries to be backend-agnostic, and do the sync work on the client side.