TopolearnCloud models let you use models too large for your hardware without changing your tools, but data leaves your machine.
Some open models are far too large for a laptop. Ollama cloud models run them on Ollama's servers, while you keep using the same app, CLI and API.
From the app or CLI: sign in with ollama signin (or the app's setup prompts), then run a cloud model by its :cloud name, for example ollama run gemma4:cloud. Your local Ollama forwards the request to the cloud. The model list at ollama.com (filter by cloud) shows what is available.
From code, directly: create an API key in your ollama.com account settings and set it as the OLLAMA_API_KEY environment variable. Then call https://ollama.com/api/chat with the header Authorization: Bearer <key>, or use https://ollama.com/v1 with OpenAI-compatible clients. With the API, use the model's full name from the list (such as gemma4:31b). In Python, Client(host="https://ollama.com", headers={"Authorization": "Bearer " + key}) works with the same methods.
Trade-offs:
- Data leaves your machine. Ollama states that cloud prompts and responses are only used to serve your requests and not for training; still check whether that meets your requirements.
- Usage is metered per account, and models can be retired; your account page shows usage and upcoming retirements. Check current plans on ollama.com rather than relying on remembered prices.
- Some features differ: for example, the docs say structured outputs aren't supported in the cloud, and stateful parts of the Responses API aren't available.
To guarantee nothing goes to the cloud (for example on company laptops), disable cloud features with OLLAMA_NO_CLOUD=1 or "disable_ollama_cloud": true in ~/.ollama/server.json.
Example
# CLI: sign in once, then use ":cloud" model names
ollama signin
ollama run gemma4:cloud "Explain vector databases in two sentences."
# API: call the cloud directly with an API key from ollama.com
export OLLAMA_API_KEY="..."
curl https://ollama.com/api/chat \
-H "Authorization: Bearer $OLLAMA_API_KEY" \
-d '{"model": "gemma4:31b", "messages": [{"role": "user", "content": "Hi"}], "stream": false}'
# Python (same library, different host)
# from ollama import Client
# client = Client(host="https://ollama.com",
# headers={"Authorization": "Bearer " + os.environ["OLLAMA_API_KEY"]})
# Company laptops: keep everything local
OLLAMA_NO_CLOUD=1 ollama serveRemember
ollama signin+:cloudnames in the app/CLI- Direct API: https://ollama.com with
OLLAMA_API_KEYas a bearer token - Data leaves your machine; some features differ;
OLLAMA_NO_CLOUD=1disables it
Watch out for
- Assuming every local feature works the same in the cloud, such as schema-constrained structured outputs.