I released first version of the decoder, with the databases. It decodes english texts well; russian texts are tricky, though.
Use the big english database - it contains about 128000 words and has a good perfomance - can decode even short texts (see 'john.coded' example included in sources - only 31 words).
Encoder can cipher text by replacing letters with other letters in the whole text. For example, it can encode the first sentence of this text into this: "Koajrkz aeo asxlkz pkdp tv zkxqeasoi qkppkzw fspl jplkz qkppkzw so plk fljqk pkdp.". I'm now writing the decoder, which will hopefully decode texts back into English (or other language), without the knowledge of which letters were replaced.