Thank you! Very interesting. When building a sequential agent that would work on a very large document as input, and iterate multiple steps to generate another large document (some kind of report), is there some kind of good practice to store the generated document at each step? Especially if the document is too large to be sent entirely at each loop (let alone sending the full history to the LLM), maybe having to work with chunks. And maybe having a human in the loop to review the generated document multiple times during the process. Some kind of 'large artefact'. I'm not sure if the SQLlite memory node is appropriate. (Also, I guess the SERPAPI node would be replaced with some sort of document loader but I cannot get the "Read File" tool node to work with a sequential agent). Regards.
Great video, thank you very much. Do you have any experiences regarding the performance when reviewing large files?
Thank you! Very interesting. When building a sequential agent that would work on a very large document as input, and iterate multiple steps to generate another large document (some kind of report), is there some kind of good practice to store the generated document at each step? Especially if the document is too large to be sent entirely at each loop (let alone sending the full history to the LLM), maybe having to work with chunks. And maybe having a human in the loop to review the generated document multiple times during the process. Some kind of 'large artefact'. I'm not sure if the SQLlite memory node is appropriate. (Also, I guess the SERPAPI node would be replaced with some sort of document loader but I cannot get the "Read File" tool node to work with a sequential agent). Regards.
I have the error "message stopped", what is the posible cause? I updated all dependencies.