I keep working on knowledge capture tools at home, off and on. I want to be able to use the computer as an aid to thinking, but so often the tedium and hum drum nature of the thing limit me from using the computer. Using the computer as an advanced notepad only goes so far: OneNote is nice, but it's incredibly proprietary; OrgMode is crabbed; mind-maps are graphical and too specific for a single purpose. I've tried mediawiki, and I lost that years ago because MySQL is a binary format and I didn't migrate the data.
Things like metadata matter; yet, not only does it matter, metadata is hard to keep up to date. So part of the work has been to build a content-indexer for textfiles.
I'm working on an annotator (applied graph database, basically) system now. Automated categorization derived from the indexer will be fed into the annotator, then used to inform on the files. Ideally, given this automated information, I can manually add useful tags to a file, then derive similarity metrics from one file to another.
That is all to say: I want to build an automated cross-referencing tool so that when I write down thoughts, I can find them again, years later, without having to do significant manual tagging work. Speaking of tagging - I'm not using single tags for this system: that's silly. Hierarchy is a thing and a very useful way of organizing information. But, having only one hierarchy is very limiting, because taxonomy is difficult to correctly perform. So I'll be lazy: a file (a knowledge bit), will have multiple hierarchies in the database, as many as makes sense. From a file system perspective, this turns the whole idea of a root filesystem on its head: now every file is a root that many paths can go to.
Another thing I'm working on slowly is a better interactive mode for the shell. Command line is incredibly powerful, but generally it's been left in roughly 1989 or so; bash is an awful system, and horribly tied to Unix 1.0 design. One thing I particularly want to be looking at sorting out is long-term storage of web pages. Many very very interesting pages have gone into /dev/null - and those are the ones I've read! I've dabbled in the Xanadu demo from Ted Nelson, and the concept of persistent transclusion without linkrot is powerful; I want to achieve some of that with this.
The aim here is to let me think and connect thoughts to form very deep thoughts with complex ideations involved in the process. In a sense, it will be a memory system customized around how I want to think.