Yesterday I posted about the packages I was using with Pi coding agent. Someone asked if I was running pi against local models.
My answer turned into a blog post, because there's a LOT of misinformation out there about using LLMs with local models.
short version: unless you're rich, it's only technically possible, not practical. However Pi is probably the best tool out there for minimizing token usage on local or remote models.
https://weblog.masukomi.org/posts/the_thing_about_local_llms/
