I like Romans post near the bottom - every time I read I find lots more things to try out.
I don't think that anybody has completely nailed generative algo's so I think it is still a bit of a fiddle about and see what you can get going. Probability, Scaling, and Markov chains are some of the processes typically used.
to export midi you need to record it first- use [seq ] for that. to apply a seive you could look at coll as an intermediary ie random 8-> coll with 8 indices as addresses for degrees of a scale; ie 0, 60; 1, 62; etc
I guess if you would like to do algorithmic composition, you'll need to learn programming at least on some basic level. Otherwise you won't be able to tell a machine the algorithms you invented...
With that said, Max is a good environment to start with. Even if you had zero knowledge on programming, learning the basics is quite easy and straightforward (compared to, for instance, plain C or Java, or more 'code-oriented' environments, like CSound) if you just follow the tutorials, and it will take you some time to reach the limits of the system.
I've never programmed personally in OpenMusic, so I can't tell you what that could do for you. However, I have plenty of colleagues that work with OM, and most of them told me that that system is more suitable for algorithmic composition than Max. So, you might want to have a look on that environment too (if you have access to IRCAM software).
There are over one hundred documented computer programs/environments/languages intended for composers. Christopher Ariza maintains the most comprehensive, consistently documented catalog of them on the internet:
Oh.... Finally it is the same. I know all these sites and I read a lot. Just I am so curious if the Max has enough power for algorithmic composing in the matter I need. It gets again a confusion.... :(