The title may have been a little confusing but basically the memory system how you’ve got it set up how it’s saving the 1 line memories is completely awesome so the next step what I would do is because your vector file search is so awesome
I would implement that automatically into chat GPT so every time I add a nameable code item like a function which I could reference by name or an operator which I could reference by name you know any nameable code I use python I don’t know what other programming languages have but those are the two basically nameable items you could go up to working with types but for starters if you just do this (this may actually supersede or should supersede the memory system)whenever a user references a function name or operator name in their request… I would have a background assistant running on a superfast server look up that function and just add that function or operator to the assistant memory for that request so So the memory of the contents of that function or operator would be essentially immediately provided as an example and if you limit these files to around 7 megabytes you know each function or operator in a JSON format on a separate line then it would be pretty quick to access this especially if you did run this functionality from a better server or a superfast server this way whenever I mentioned to the assistant something about a function or operator they would immediately know what I’m talking about and the function or operator content now…
I know people like to voice type their user request so you could instruct people in your documentations or even making a print statement right on your web interface that when they’re referencing a function and voice typing they need to reference it by voicetyping for example print _variable _function so the actual voice typing of that would be ‘print underscore variable underscore function’ Then another assistant in your chat GPT code could be looking for any underscores and if it finds one that maybe has words that are separated by underscores like that that may have a space as sometimes happens with Windows voice typing when you tell it to print an _between the word and the _ or the _in the next word of the function name then it would find those and fix those just like it does voice typing errors…
This will be an awesome starter for saving history in an even better way between all sessions because the memories would be linked to keywords…e.g. function names And when a keyword like that is provided then the item representing that keyword would be sent to the assistant’s memory for that session… You could also save the names of recipes for Housewives like if somebody’s making a macaroni and cheese bake then the memory system could create a keyword called Macaroni and cheese bake linking the entire recipe to it and just be searching for that reference in that person’s future…requests… And also you could go the next step and add functionality in the background like how you could tell the assistant specifically save this into memory and it does you could say Link this entire request to a keyword or even give the option to specifically state link All the code I’m providing even the code that’s outside of any functions or operators to this keyword… And then it’s super simple it’s automatic but it’s also invokable like your memory system.