How can I reduce the context window?

[madorian]

If I have a discussion, I might go back and forth endlessly, so my chat buffer reaches 2000 lines. I then start quickly to get Claude limit messages.

I was thinking to ask the LLM to reduce the chat to only include what is essential, but I guess that's also a big output.

What does one usually do here?;)

[karthink]

There's no token-based truncation available in gptel. But you can limit by the number of exchanges, which is imprecise but might be enough.

Run (setq gptel-expert-commands t), then set the -n option in gptel-menu to the number of exchanges (prompt + reply) to include in the context.