Ollama cloud models: does your data leave your laptop?

Ollama cloud models are models that run on Ollama's servers instead of your computer, so yes: when you use one, your prompt leaves your laptop. Local Ollama models run on your own machine, and Ollama's FAQ says, "We don't see your prompts or data when you run locally."

Both kinds sit side by side in the same app and the same terminal. That is handy, but it also means you can send private text to a server without noticing. You will see what Ollama itself says about cloud data, how to tell the two apart, and how to switch the cloud off.

What are Ollama cloud models?

Ollama cloud models are hosted models you call through Ollama without downloading them. The official cloud page says you can "Run models in Ollama's cloud from your apps or terminal. No model or app download required."

You can reach them in two ways:

• App or CLI: sign in, then use a name with a ":cloud" tag, such as "gemma4:cloud".
• API: send requests to https://ollama.com/api/chat with an API key, using the model name from the cloud list, such as "gemma4:31b".

Cloud models are useful when a model is too big for your laptop. A student with 8 GB of RAM cannot run a 31 billion parameter model at home, but can call one in the cloud. Check the Ollama pricing page for the current free limits, since plans change.

Ollama local vs cloud models: what is the difference?

The difference between Ollama local and cloud models is where the computing happens, and so where your text goes.

• Local model: you download the file first ("ollama pull"). The model runs on your CPU or GPU. Your prompt stays on your computer.
• Cloud model: nothing big is downloaded. Your prompt and the reply travel over the internet to Ollama's servers. You need to be signed in.
• Speed: a local model is limited by your hardware. A cloud model is limited by your internet and your plan.
• Offline use: a local model works with no internet. A cloud model does not.

Our post on small language models you can run on a laptop covers which local models fit on a normal student machine.

What does Ollama say about cloud model data?

Ollama says it processes cloud prompts to answer them and does not train on them. Here are the exact words from three official pages, as of November 2026.

The cloud docs page, under "Data handling": "Ollama processes cloud prompts and responses to answer your requests. We do not use them to train models."

The FAQ: "When using cloud-hosted models, we process your prompts and responses to provide the service but do not store or log that content and never train on it."

The Privacy Policy (last updated March 2026): "When using cloud-hosted models, we process this content transiently to provide the Service and this content is not stored beyond the time required to fulfill the request." It also says, "We do not use your inputs or outputs to train any AI models or request prompt or response content in support requests."

Two more lines from the Privacy Policy are worth reading closely:

• Third parties: the policy lists service providers that help Ollama operate, "(e.g., Stripe for payments, cloud infrastructure providers, model inference providers)".
• Metadata: Ollama collects "basic account info and limited usage metadata that does not include prompt or response content".

So the fair reading is this. Ollama promises not to train on cloud prompts and not to keep them after the request. But a cloud prompt still leaves your machine, passes through Ollama's systems, and may be handled by infrastructure or inference partners. A promise in a policy is not the same as the data never leaving.

How to tell if an Ollama model runs locally

An Ollama model runs locally when you downloaded its file and its name has no ":cloud" tag. Use these checks:

1. 1Look at the name. "gemma3:1b" is local. "gemma4:cloud" is cloud. The tag after the colon tells you.
2. 2Ask whether you downloaded it. The cloud docs say, "Cloud models do not need to be downloaded." If a big model started instantly with no download, it is probably cloud.
3. 3Turn off Wi-Fi and try again. A local model keeps answering. A cloud model fails.
4. 4Check the code. If your script calls https://ollama.com/api or uses an OLLAMA_API_KEY, it is using the cloud. A local server listens on localhost, port 11434.

How to run Ollama fully offline (local only mode)

Ollama has an official local only mode that turns off cloud models. The FAQ says, "Ollama can run in local only mode by disabling Ollama's cloud features." You lose cloud models and web search, and keep everything else.

1. 1Open the file ~/.ollama/server.json (create it if it does not exist).
2. 2Add this setting: { "disable_ollama_cloud": true }
3. 3Or, instead, set the environment variable OLLAMA_NO_CLOUD=1.
4. 4Restart Ollama.
5. 5Check the logs. The FAQ says they will show "Ollama cloud disabled: true".

After that, pull the local models you need while you still have internet. Then the work can run with no connection at all.

What should students with private research data do?

Students with private research data should use local models only, and turn the cloud off so a wrong tag cannot send data out by mistake. Private data includes interview transcripts, patient or survey records with names, data under an ethics approval, and your own unpublished paper.

• Read your rules: your ethics approval or data agreement may say data must stay on approved devices. A cloud call can break that, even with a good privacy policy.
• Disable the cloud: set OLLAMA_NO_CLOUD=1 on the machine that holds the data.
• Remove names first: if you must use a cloud model, strip names, phone numbers and IDs before you paste anything.
• Write it down: note in your methods which model you used and whether it ran locally. Reviewers and supervisors will ask.
• Use the cloud for public text: summarising published papers or fixing grammar in your own draft is a lower risk.

Our view: cloud models are a fine tool, and Ollama's policy is clearer than many. The risk for students is not bad intent. It is a single ":cloud" tag typed out of habit. Local only mode removes that risk in one line. For more on keeping AI tools in safe limits, see our post on AI guardrails.

Want to learn research methods with other students who care about doing it right? Join APJ-RAISE, free for students 18+.