atyh
· 1d
interesting. do you have the time to explain the concept to me like im 5? if not i get it. this is all new to me so im wading in without knowing the lingo. plus, the key experimentation hasnt been cod...
Agents need a way to 'remember' what was concluded decided 4 chats ago. It can't just load up the entire previous conversation into the next chat, that would just eat up the context window all over again. So a memory system gives an agent a tool to look up memory worthy artifacts. For example when I mention a specific machine, I shouldnt have to remind the agent what IP it is, if it has ssh key access, what OS it is, every single chat. This facts become addressable artifacts that are quickly injected into context as needed.