[Feature Request]: Memory

[NulledInstance]

Description

add a memory system that makes the ai summarize the last few (or more) messages before putting it into a list, and then after it should discard the summarized messages out of the context window so that offline running llms doesn't take a long time. this can probably help with continuing chats after the model is unloaded since it doesn't have to reread the past messages again, only the summarizations. this can also help with using short context window llms or making it remember longer

Proposed Solution

No response

[tulang3587]

This would be a great addition. I'm thinking if we can determine the prompt to get the summarizer to generate the summary for the memory, and then limits how many chat messages that are included in a chat (while also including the summary) would be great.

I've been facing some issues where if the conversation gets too long, the model gets confused and started returning gibberish.
I had to generate a summary, then in a cloned chat list, I edited the first messages with the summary and deleted most of the conversation log in order to continue.

Please add this feature 😊

[spirited-wind]

I also think this feature would be very helpful.