Context usage going exponential for some reason, wierd claude code internal tracking? or are we missing to compress some request? potentially haiku classifications? Should be compressed butm hmm warning will fuck up your usage after a while , the tokens we send to the api are so few, and thought it could be context caching problems but no it goes exponential in usage! either we somhow send 500k tokens t a model that dont suport it??? or idk man.
and it takes it or claude code tracks the usage internally and we have to intercept that? but htat is stupid
Context usage going exponential for some reason, wierd claude code internal tracking? or are we missing to compress some request? potentially haiku classifications? Should be compressed butm hmm warning will fuck up your usage after a while , the tokens we send to the api are so few, and thought it could be context caching problems but no it goes exponential in usage! either we somhow send 500k tokens t a model that dont suport it??? or idk man.
and it takes it or claude code tracks the usage internally and we have to intercept that? but htat is stupid