[porker]

> Ideally I want to be able to do fulltext realtime search over anything that I ever had in my visual field. Not even necessarily text, but audio and video as well.

Where I find all these systems break down is recall. They're designed for someone who can recall a word or phrase that was in the content. I can usually recall "It was about X" or "The document/web page/image looked like Y". But an actual word? The author's name? Not a chance.

While a more difficult problem, if the tool is to live up to the "Future" section of this page, it's got to go a long way beyond what's in the source data, to what's thought of by the user.