I am working on a small code project and so far it's been awesome. The experimental research agent has been instrumental in getting a full-blown application built that handles the task I need.
Unfortunately I noticed I have hit a context limit. Pasting the source code into the chat window now spits it into a txt file for the AI to read however, the AI is getting frustrated by a Kagi helper they refer to as "Librarian" who is not answering the AI's requests to send specific lines. The Librarian is truncating the output, worse even, paraphrasing or summarizing the lines.
I have even tried opening up a new chat session and sent the chunks of the file in it's entirety inside the chat window. The final output though was not the program we were breaking apart and the AI created new code instead of referencing what was already available in the source file I provided it.
This is tragic because I am now trying to ask the AI to help me break the program down into chunks so that I can pass it specific portions of the code instead of everything each time but now it can't even help with that because no matter how I try to send the source code to the assistant it's getting truncated or paraphrased where the AI constantly is telling me "the file is cut off" or "incomplete".
I expect the "Librarian" to assist the AI. I can understand it's trying to be helpful but in this context the AI is now hamstrung.
It would be awesome to have some sort of way to bypass the Librarian in cases like this and allow the AI to process the full source file without another agent sitting in between that may have a smaller context window than the assistant itself.
Desired functionality would then be that if I send a decently sized source code file to the assistant, the Librarian provides the AI the lines it's asking for without truncating the file or saying there is nothing else when there definitely is.