The receipt that separates local from cloud tab.
Step 1 — Turn off Wi‑Fi
- Laptop: disable Wi‑Fi (airplane mode is fine if that's easier).
- Desktop: unplug Ethernet or disable the adapter.
Don't close Ollama — if it's running from earlier, leave it.
Step 2 — Run your model again
Open terminal and run your tag:
ollama run llama3.2
(Use your model name from ollama list.)
Step 3 — Ask something simple
What is 17 × 23? Reply in one short sentence.
Wait for the answer. It may be slow on CPU — that's normal.
Step 4 — Turn Wi‑Fi back on
When you're satisfied, re-enable network.
You should see: a sensible answer while offline — proof inference ran on your machine, not a remote API.
If it fails offline
| Problem | Likely cause |
|---|---|
"connection" error on ollama run | Ollama service stopped — restart tray app or ollama serve |
| Hangs forever | CPU-only first token — wait 30–60s once |
| Can't run at all offline | Model not fully pulled — redo Lesson 3 pull |
I still run this exact test after every reinstall. Boring maths prompt, Wi‑Fi off — if it answers, I trust the stack again. It takes ten seconds and saves a lot of second-guessing.
Continue — what's next.