Make the save on ur own side, use resume and recursive injection in the prompt each message, not too complicated, ask for it to the damn bot, explain it carrefully. Dont go over the 6k token limit and u will be all good.

you don't, you can only be creative about ways of recalling info while relying on summaries, the agent before it got replaced by muse spark ideated two systems, one named "hard facts EAV" that analyzes the story and grounds hard facts that shouldn't dissapear or at least last more, then a "bm25" that can retrieve forgotten paragraphs by similarities but you're on a tight budget anyways, context is 6k, it's theorically 8k but 1024 are reserved for output

There is tricks I don't know about to make the most of what we've got, but the short answer is use a more expensive model than the one on Perchance. I don't want to be rude, but this is like asking how to generate 2048x2048 images with the text-to-image plug-in.