Its implementation may be a bit academic for what you're looking for, but Watson Beat (https://github.com/cognitive-catalyst/watson-beat) can generate music given a seed melody a number of configurable parameters.

All output is in MIDI files, and by default need to be glued together manually (though I suspect this could be automated).

I've also wondered how interesting this could get if it could be configured to dynamically feed the generated melodies back in as the seed, effectively evolving the music over time while remaining somewhat familiar from one adaptation to the next.