I'm on the verge of getting washed away by cool things again. The same guy who pointed me towards CRM has now pointed me something similar but in many ways superior - POPFile. This is another Bayesian filter dealio, but this one is a proxy through which you run your mail. This proxy maintains a history of what has run through it, and has a built-in web server that provides a GUI to manage training sets. It also allows for multiple buckets, so you can train the spam, work, personal, or whatever matters to you. It will add a header to all mail that runs through it, so your mail clients could filter based on the classification that POPFilter did. One of the big problems is that when you set up filters on Outlook or Eudora, none of that is portable. This way, the classification exists in your mail no matter what you use to read it. Cool idea. At the moment, I am now drowning in Bayesian stuff.