Auto-trim content if too large for context window

[xpluscal]

As far as I can see there is currently no native way here to ensure we don't hit the max context window?
Would be awesome to have base agents use tiktoken or similar to handle this - or am I missing something?

[KennyVaneetvelde]

Nope you are correct this is one area I have slightly neglected. I am open to suggestions on implementation here, tiktoken can be used for openAI, but we'll need to have an elegant way to support it for every possible provider or have a nice way that we can chip away at it in an extensible way