When does Ollama keep the work on your machine?

Select a supported local model and leave Ollama’s cloud features off. Ollama says it does not see prompts or data when a model runs locally. Its documented local-only mode disables cloud models and web search.

Ollama documents that its local endpoint does not need an API key. Keep a practice copy private and use dummy details. A live app needs its own access checks and limits before letting visitors use a model.

What changes with the selected model
Choice	What happens
Local model	The runner loads files and uses working memory on the machine. Your app can still save logs or send information elsewhere if you connect it to another service.
Cloud model	The hosted service processes the prompt to return an answer. Ollama says it does not log or train on prompt and response content and says it does collect some account and usage information.

How do you choose a first example that fits your computer?

Start with a short piece of text and a task you can judge. You can plan the example on paper without installing anything. Check the runner’s supported hardware and the size of the exact model you would choose before trying it locally.

Write two invented order facts
Use “Pickup Thursday” and “Bring your receipt.” Ask for a two-line pickup reminder.
Check the result you need
The reply should keep both facts and leave out an address or order number. Keep this example outside the model runner.
Check the size before trying it on a machine
Look at disk space for the model files and working memory when they load. Longer requests can need more memory. If the machine cannot comfortably run the model, consider a smaller compatible one.

What changes on your bill if you switch to Ollama Cloud?

Local inference uses your hardware. Ollama Cloud charges by model usage. Its pay-as-you-go option has no subscription fee and still meters cloud requests.

Cloud pricing can distinguish the input, cached input and output pieces the model counts as tokens. Paid plans may include a set of credits with a monthly reset. Check what happens when those credits run out, whether any automatic billing is enabled and what happens if you send more requests at once than the plan supports.

Can you take the model and settings to another machine?

Keep a copy or a release reference for the model files you are allowed to use, plus the settings your app relies on. Check whether the destination runner supports the exact model format and architecture.

Ollama’s MIT license covers the runner. Models have their own terms. Keep app data, expected replies and business rules in ordinary files you control. Moving the runner or model files does not carry over chats, app records or an Ollama Cloud plan.

What can you check when you change a local model?

Repeat the small pickup example with the same two facts. Compare the response with the expected reminder and notice if either fact disappears or changes.

Record the model and settings that answered so you can repeat the check. If the new model does not fit or produces an unusable reply, keep the simple pickup instructions available to the reader while you choose another setup.