It's case sensitive and has related problems in finding words. Entering all lowercase letters recommended.

While I'm sure it was a fun exercise to build it and there will most likely even be "helper apps" appearing on the App Store soon, I frown upon these. Letterpress will lose a lot of its fun once cheating takes over.

I was thinking that an app where you could take a picture of letterpress screen and it would use that and with some strategy (in terms of defending letters) would be fun project to make. Maybe more than playing the game :)

I have done that with Spelltower (: https://github.com/flixic/spelltower

From my experience cheating, it would help to restrict words to 7 letters. Then the blocking rules would be more strategic, and at least the AI would be more fun to make.

Yes, but it would still be fairly easy. My AI is already not trying to use the longest word, but instead tries to secure as many tiles as possible. In many cases it means using shorter words that have just the right letters.

I think the solution is just to only play with friends who you know aren't cheating.

Implementing a Letterpress cheater in Python took me around 10 minutes. Computers are very fast nowadays: https://gist.github.com/3968275

There is no need for a whole game engine if you want nothing more than to know the best possible word to play each turn. It will be enough to win with 99% of people. If you want to win a Letterpress AI tournament, you would of course need to look at alpha-beta and other algorithms.

I made my own primitive helper. http://inarticulate.eu/letterpress

Neat. I'm working on JS-based word suggestion (=cheater) for Letterpress, and there were a few bits from here that I might borrow.

I'd love to hack on this to extend it; do you plan on writing any tests for it?