this script tracks three things. user facts like name, likes, fears, and goals get auto extracted from what you type. story moments with salience scores remember big emotional events. narrative threads track ongoing plotlines that are still active. basically, giving it semantic and episodic memory. it also watches how you talk and builds a behavior profile. it can tell if you are often apologetic, curious, making promises, opening up, or using pet names. every message it injects a memory block into the character scenario with the most relevant info. it cleans out old blocks first so they dont pile up. it also has hard character limits so it doesnt eat your token budget. when memory gets full it prunes the lowest priority facts instead of just the oldest ones. it can also summarize similar facts, like squishing seven likes into a short list. you can manually add memories with remember colon something in brackets. everything is tweakable through the config section at the top. inspired by @soradellam's lightweight token cleaner (remeber that these scripts are FUNDAMENTALLY DIFFERENT). use her script if you just want the bot to remember main plot points without the extra stuff. use this script if you want the bot to actually know you (self proclaimed).

DO NOT CHAT WITH THE BOT, CLONE THE LOREBOOK