I don't want to send my data to third party vendors all the time. But from my experience, the LLMs needs to be quite beefy in order to understand tool-calling, especially at longer contexts (200k+).

Before I dive headlong into investigating this and spend money on a project doomed to fail, do anyone have experience with a local model which can handle this sort of workload? I intend to run it on decent gaming CPU with 64-128GB ram.