I've skimmed through the source code, so I'll save y'all some time. It's all based on a PRNG, which is based on an AES-256 in CTR mode (making it a stream ciph

I've skimmed through the source code, so I'll save y'all some time. It's all based on a PRNG, which is based on an AES-256 in CTR mode (making it a stream cipher) seeded with a key generated by 4096 iterations of PBKDF2-SHA256. PRNG just decrypts a stream of zeroes with a KDF-generated key. All standard library algorithms, nothing custom. To generate a password, it builds an alphabet, picks a random character from it by reading a byte from the PRNG. It is looping until it reads a byte is less than `255 - (255 % max)` where `max` is length of the alphabet, then returns `b / (255 / max)` where `b` is the byte from the RNG. Basically, dividing [0-254] range (255 is never accepted) into equal-length `max` pieces and figuring out where it falls. The generated password is checked for compliance (i.e. that randomness had produced a special character, etc.) and returned if it's good, otherwise it loops and tries to generate a password again. Given that RNG is fully deterministic, this produces consistent results. I haven't bothered looking into keypair generation - I just woke up, sipping on my "morning" coffee as I'm typing this, so I'm kinda lazy.