Ask HN: State of the art with local LLMs and agents

I like Claude Code for what it is, but I want an agentic coding setup that provides much stronger security and privacy guarantees.

What is the state of the art right now regarding running local LLMS and connecting local agents to them?

Which models are suitable and good, what hardware (at a minimum) and software are required, and which open source agents are good to drive those models?