You already mapped the rent problem in Ground Zero. Local open-weight is the other category entirely — not automatically smarter, but genuinely yours.
Same comparison, one new column
| Rented cloud (Ground Zero) | Local open-weight (Off the Grid) | |
|---|---|---|
| Where it runs | Their servers | Your CPU/GPU/RAM |
| After you stop paying | Access stops | Model files stay on your disk |
| Privacy | Their retention rules | Stays on your machine for local chat |
| Internet | Usually every message | Only for the first download |
| Quality ceiling | Frontier models on tap | Depends on model size + your hardware |
Local isn't a religion, and I'm not going to pretend it is. I still use cloud AI for genuinely hard problems. But for repeat drafts, private notes, and no meter running in the background, local is the door Ground Zero kept describing — and Off the Grid is where you open it.
What "open-weight" means here (one sentence)
Some companies publish model weights you can download; Ollama is the app that runs them. Meta's Llama and OpenAI's ChatGPT are different beasts — one you can pull onto your disk, one you rent by the month.
Why I split Off the Grid into four lessons
| Lesson | The job |
|---|---|
| Lesson 1 (today) | Pick your model, and understand why |
| Lesson 2 | Install Ollama |
| Lesson 3 | Pull it, chat with it, prove it's on disk |
| Lesson 4 | Offline test + free path complete |
Each one ends with something you can actually check — same rhythm as Ground Zero.
The moment local genuinely clicked for me wasn't reading a blog post. It was running the exact same Example Ink prompt from the comparison lab and realising the reply never left the building. Lesson 3 is that moment for you. Today is just the map.
Continue — pick your starter model.