Need to do some clustering on high-dimensional datasets? If you're on Linux (or FreeBSD with Linux binary support) you could do a heck of a lot worse than George Karypis' Cluto package.
Clusters feature vectors, clusters similarity graphs, has 7 different clustering methods, 4 different similarity criterion, 12 different clustering criterion, and all of the standard scaling methods on the features/similarity.
And it generates pretty pictures too.
What more could you want?!
Okay, yes it comes with a static library so you can wrap your favorite scripting language around it. Short of being open source, it's quite the bee's knees.