RAG should be as efficient as possible. On Windows, when given a folder with several files and subfolders, As a default user option, ChatRTX should first parse out likely search terms, feed each of those to the Windows Search API, then internally reissue the query prioritizing the file names returned from Windows Search. This will speed up results dramatically. The current implementation is very slow when querying across many files and folders. This, of course, assumes the targeted folder and all subfolders are indexed.
Example. I have a target STUDIO folder. Within it are many subfolders with manufacturer names, each containing audio plugin manual PDF files. One of the last folders is Yum Audio, a manufacturer of a product called Crispy Clip, for which there is a PDF manual. A Windows search for just the term “Crispy” shows the correct file within the the top three within milliseconds. Contrarily, I let ChatRTX run on the query for “What does Crispy Clip do?” for 45 minutes without any results before stopping it. If ChatRTX were to simply pick out the parts of speech, even if it’s just to pick out the main subject of the query, and utilize Windows Search API for prioritizing the order of its file processing, results could be returned much sooner. Again, this would be an optional UI setting on by default. If the user wanted ChatRTX to process without utilizing Windows Search, they would uncheck the box.