How is the memory handled?

[sagarspatil]

Hello,

How does the memory feature work? I know the data is stored locally, but do you embed the conversations and share them with the model? How can it refer to a conversation say that is beyond 8000 tokens? (In case of GPT-4) or will it lose context?

[ztjhz]

It shares the entire conversation to the model. If it exceeds 8000 tokens (for GPT-4), the conversation will be truncated to less than 8000 tokens.

[jonmatthis]

I was thinking about integrating some of the LangChain memory management methods into this (lovely lovely) project

Does that align with your roadmap? If so, I'd love make a PR back into this branch when I've got it working!

[jonmatthis]

LangChain is Python based, so integrating it might be a bit tricky, but I think it would add a lot of potential applications to this project.

This project implements a nice combo of ReactJS frontend code with Python based backend code, with communication managed through a docker-compose.yml that makes containers for each one - https://github.com/logspace-ai/langflow

[jonmatthis]

Anyway, let me know if y'all have any plans for handling Memory in a more sophisticated way - if not, I'll just poke around and let you know if I figure out anything useful :)

It might take the form of a PR into your repo, or maybe just an integration with some other codebase that I will make on own (if it looks like it will be hard to do with out making a lot of deep changes to your repo)

[ztjhz]

We do plan to handle memory in a more sophisticated way!