I gave my local LLM persistent context, and it finally stopped making the same mistakes

It's not memory, but it's close enough