NovelAI uses something like this with their home-rolled models: you can put in some generic 'memory' as well as keyword-based notes and it will automatically ge

NovelAI uses something like this with their home-rolled models: you can put in some generic 'memory' as well as keyword-based notes and it will automatically get included up to a certain token limit, including some prioritization handling for cases where many different notes are triggered at once.