Devlog #1: Kicking off the project. I started off with the idea of having all L.L.M.s work inside it’s specific folder and writing down memories saved themselves. Yeah, that’s a bad idea. They’re stochastic. They hallucinate. Unreliable. So I changed my plan: instead of having them write down misinformation, why not “interrogate” their terminals myself and let the actual machine do some of the work? Using FTS5 and SQLite along with a bunch of .md files, Noesis will allow for cross-L.L.M.s collaboration and a much bigger memory. Not much has been implemented so far tho… but it’s just the beginning! 🐱

P.S.: inspired by Claudian (https://github.com/YishenTu/claudian)