(Not sure if the source code for this will be released currently.)
Run with (to train the model): python2 model.py
cIon utilizes a embedded recurrent neural network to allow for natural conversations to take place.
A dataset (16.3MB) of a little over 300,000 lines of movie dialogue was used.
Step time is about: 4.41 seconds
It has been trained for just about 15,000 steps so far within 24 hours over the course of three days.
(Because I didn't run it all day long.)
cIon seems to reply quite well at it's current stage.
Examples: (These aren't hand picked, they are presented as they happened on the 14,400th step)

Me: hi.
cIon: hi.
Me: hello
cIon: hi.
Me: how are you?
cIon: i'm fine.
Me: what's up?
cIon: nothing.
Me: what is the meaning of life?
cIon: the people are.
Me: should i kill someone?
cIon: no.
Me: what is your favorite thing to do?
cIon: i don't know.
Me: have you had a girlfriend?
cIon: no.
Me: do you want a girlfriend?
cIon: no.
Me: why?
cIon: because I don't know.
Me: does god exist?
cIon: no.

The bot seems to have some morality.
It's context understanding is debatable at this stage but from previous conversations I've had with it, it did demonstrate basic context understanding.
Note: Not all of it's sentences are so short.
This is simply a quick convo I just had with it.