The publication for agent-led work.

Run AI on Your Own Terms: The Complete Local Inference Guide for 2026 — From Zero to Hardened Server

Local AI inference has crossed the threshold that matters: a 284‑billion‑parameter frontier model now runs at 26 tokens per second on a MacBook Pro. But the explosion of self‑hosted LLMs has been met with 175,000 exposed Ollama servers, 91,000 documented attack sessions, and a critical memory‑leak vulnerability — "Bleeding Llama" — that leaked heap memory from roughly 300,000 internet‑facing instances. This guide gives you the hardware picks, the runtime decision matrix, the Nginx config, the firewall rules, the pre‑flight checklist, and — most importantly — a step‑by‑step walkthrough so you can go from zero to a secure, production‑grade local inference stack in an afternoon.