Question Persisting information in models

Hey everyone, I'm new to ML agents and the forums so apologies in advance if this question has been asked and/or answered before, but is there a way for your agent to keep information between decisions? More specifically, is it sufficient to just have my pytorch model keep the desired information in variables, or is there more to do on the unity side to ensure the information isn't lost after a decision is made? For example, the hallway agents clearly have a way to recollect which shape they saw before heading to the goal area, but it's hard for me to tell how this is done since the model is a black box.

Also on a slightly unrelated note, is it possible to get the model's weights and layer values after forward passing?