A way to break down what's in the context

[lockie]

To upvote this issue, give it a thumbs up. See this list for the most upvoted issues.

It'd be great to have some kind of built-in observability tools which clearly show what is the current content of the context window, what was sent previously and what will be sent. Kind of like Claude's /context command.

[ericdallo]

Sure, the problem with that is that each model/provider count differently tokens, right now we rely only on the post request, when the provider said how much we used, this is assertive, otherwise we would need to have a way to genericly count tokens or specific by provider, which something I'm considering adding support somehow.

[lockie]

Nice. To cover 80% of this, the exact token counts are not needed I think, just the files/pieces of text representation, perhaps with just indicated word count (which still gives a good approximation of how much tokens there will be).