muscle mem

+1 on the idea

basically cache the tool calling

but i suspect this kind of cache can be easily enabled in llm api caching functionalities or agent framework level

this is caching on the llm api level

I think a better "caching" is condense the experience / knowledge into data / code.