Pass this checkpoint if all three are true:
ollama --versionprints a version — no "command not found".ollama listshows at least one model you pulled.ollama run <your-model>answered the Example Ink prompt with three post ideas.
If all three pass: you just ran open-weight AI on your own hardware. One more free lesson — the offline proof — and Off the Grid is complete.
Common fixes
| Problem | Fix |
|---|---|
ollama: command not found | Reopen the terminal after install; Windows: reboot once; confirm install from ollama.com/download |
connection refused on pull or run | Start Ollama (tray/menu-bar app, or ollama serve / sudo systemctl start ollama on Linux) |
| Model name not found | Run ollama list — use the exact NAME column; re-pull if it's empty |
| Reply is gibberish or empty | Wait longer on the first token (CPU-only machines are slow); try a simpler prompt: Say hello in one sentence. |
| Machine freezes or kills the process | Model's too big for your RAM — remove it with ollama rm <name> and pull llama3.2 instead |
| Download stuck at 0% | Check your firewall/VPN; retry ollama pull |
Still stuck after two fixes? Note your OS, your RAM, and the exact error text — the later lessons assume this step works, so it's worth getting help before you move on.
My own first local run was CPU-only — a full twenty seconds before the first word appeared, and I genuinely thought it had broken. It hadn't. Be patient on that first token; after it, the text streams like anything else.
Continue when all three checks pass (or you've sorted the issue).