What a great confirmation of what I've tried to achieve by getting models to communicate with my local data in shorter communicates to save on token command bandwidth, more or less the same results! Thanks for the article!