Nice setup. Running models locally on Mac hardware has gotten surprisingly viable. I'm using a similar stack in Switzerland for testing AI agent workflows — the M-series chips handle inference well for tool-calling tasks.