The context window of calls currently supports none, all, last n, or first n. This may work well for short or very narrow conversations, but in exploratory conversations where you are learning new things about a broad topic, it is very easy for context to be lost with this model. To be more dynamic without losing significant context of longer conversations while also being more cost effective than the "all" option, it would be nice if there was an option to let it sent a request in the background to have the LLM summarize the conversation history so far and use that response as the context moving forward and then repeatedly do this from the last summarize request (+n messages) as the conversation gets longer.

This is similar to how LLM assistants work in text editors. In fact, here is the one that VSCode's GitHub Copilot Chat uses to summarize the conversation when a thread gets long enough:

If this feature were to be added, it would be nice to see some sort of indicator that conversation summarization happened at a certain point in the chat, as well as the ability to customize this summarization prompt in settings (a code-specific summarizer prompt might not serve as a good "general conversation" summary generator, and vice versa)