This is our first attempt to construct research (RL) agents. These are (hierarchically) driven by simple (hierarchical) entropy goals that allow the agent to be each time more informed about relations between concepts of some information input. Intially such an input is a collections of text documents.