My favorite Hypertext tool right now is Tinderbox [http://eastgate.com/Tinderbox/]. Why Tinderbox? Well, many reasons. First, the basic data model is very good: Nodes are typed, have a separate text and content, and arbitrary attributes; links are typed and can be attached either to a node or start from an anchor in the text content.
There are limitations: No attributes can be attached to links, notes cannot be tied to links, there are no destination anchors. But I can live within those limitations, which I think are fairly well balanced.
Second, usability of Tinderbox is very much done right, with more than one way to draw a link, or set an attribute. Again, there are limitations, due mostly to the OS9 roots of the code. I expect this to evolve over time. But overall a joy to use.
Third, there is a really nice system for simple automated queries, called agents. I can see at a glance that one of my articles is still not attached to a theme, for example.
Most important for me, Tinderbox data is pretty straightforward XML. Not quite the XML format I would have designed, but it's XML so I can play. And play I did; have a look at my TinderToolBox [/tinderbox/tindertoolbox.html] and other experiments [/tinderbox/].
Oh, and finally: The Eastgate team, headed by Mark Bernstein [http://markbernstein.org/], is a great team, and are open to suggestions.
This site is designed in Tinderbox, and XSLT'ed into XHTML. (Tinderbox has a very nice export to HTML system, but there are a few twists and turns in my site that it cannot quite handle, mostly in terms of the multilingualism.) Also, the same Tinderbox document is separately XSLT'd into GXL [http://www.gupro.de/GXL/], then converted to the GraphViz [http://www.pixelglow.com/graphviz/] format, and this is how the site map is generated. Très cool, if I may say so myself.
Not as snazzy, but more important to me: I am doing argumentation diagrams for my M.Sc. using Tinderbox. This is why I developed all this machinery in the first place.