It's an RDF graph mapped to a key-value database, so you describe edges in a tuple {subject-predicate-object} such as "tom-likes-frank" and while the complexity you can model depends on how you define your schema, RDF is good enough for most applications. However it isn't easy to define weighted graphs -afaik- without creating numerous predicates for discrete intervals.

I'm pretty positive about this project, it's like google cayley if it hadn't have got bogged down in trying to support numerous database drivers. It'll show its strength if and when it implements raft and reaches a stable production state.