Continue conversation and memory

[jomagalo]

This app is undoubtedly wonderful, but I don't understand why they don't take advantage of the memory available to clients like lmarena and qwen. Sending the entire conversation history with each prompt, as some chat clients do, is a huge waste of tokens, especially when there are ways to control this from the bridge.

The idea is simple, and I imagine many have already considered it. If a chat client like Msty Studio, VS Code, etc., is used, the client type could be detected through headers or some other method. For example, in VS Code, I have an extension that fully utilizes its native chat to connect with any provider. This extension, among other things, sends certain headers like session-id, conversation-id, etc., which help maintain conversation continuity by leveraging providers/models that have memory. The application/bridge should know when to deduple and/or send only the user's last message, not the entire conversation history. This achieves several things, including not overloading providers by sending the entire history of a conversation with each interaction, saving a large number of tokens, and making better use of the AI ​​model's memory.

[hlohaus]

Hello @jomagalo,

We currently support conversation and memory across all providers. For examples, please refer to our Context Management documentation: https://g4f.dev/docs/review/