Free and offline. No API bills, no token-meter anxiety, no data leaving your machine. For heavy daily use, local wins on cost alone: a local model costs whatever the electricity costs. - The offline part is underrated: internal documents, client work, regulated data stay on your hardware. That is hygiene, not paranoia. - What you give up: the frontier. The biggest API models still lead on hard reasoning, and a local model is a snapshot. The gap shrinks every release cycle. - Start with a 7B-class model on hardware you own. Scale up only when you can name the wall you hit. Most people hit it later than they expect. - Skip the terminal if you want: PocketPal AI gives genuinely free on-device chat with GGUF models. Still offline. Still yours. Want it running without the setup? PrivateLLM deploy puts a private LLM on your AWS for $50 plus usage.