Seriously, try Tinfoil

I’ve been using Tinfoil + DeepSeek 4.1 Flash for inference for a few weeks now. I’m using it for my personal agent harness, some OpenCode tasks, etc.

It works really well! Totally private, inexpensive, and capable!

If you’ve got the wherewithal to build your own tools like this, check it out.

My current mental model:

Local LLMs -> Best
Private LLM -> Next best