Keep your tools. Add the routing underneath.
Haybox speaks the standard local-endpoint protocols your tools already understand. One configuration line points your tool at the daemon; from then on, every request is classified, routed (or shadowed), verified where possible, and written to the Ledger. Nothing about your editor, your keybindings, or your muscle memory changes. Pick your tool:
A llama.cpp server works as a local tier without a page of its own, and anything else that speaks the standard protocols probably works too; the docs cover the generic setup, and the Discord knows the edge cases.
See what your own traffic says.
Shadow Mode reports on the traffic you already have. Before that exists for you, the calculator will estimate it from your GPU and your monthly spend, and show you every assumption it used to get there.