I like the idea of keeping HTML's document-centric original design, but accessing the documents using p2p protocols (instead of the client-server model used on the web).

I'm working on an open-source implementation of this idea at https://www.hyperhyperspace.org