Daniel Nguyen I don't know the feasibility of doing it, but maybe integrating a "notes" section that the LLMs with function calling are able to use could be helpful. We could then add instructions to the model to utilize this feature.