Move from Ollama Cloud to Ollama

Take it out

1. 1

   Step 1

   Read the bill, not a download button

   The pricing page checked November 12, 2026 lists Pay as you go at $0 and says you pay per token for every cloud model. The usage FAQ says running models on your own hardware is always unlimited. The August 31, 2026 post still describes a free plan with a small monthly allowance for a set of starter models. Neither page prints the day that allowance ended, and neither page describes an export of chats or of the credit balance.

2. 2

   Step 2

   Write down the model names you still call

   The pricing table uses names such as gemma4, gpt-oss:20b, and gpt-oss:120b. Copy the names your scripts already send. A local pull uses those library names. It does not copy the hosted weights onto your disk for you.

3. 3

   Step 3

   Leave the cloud key in place until a local reply works

   A script pointed at the hosted API keeps billing until you change it. Do not delete the key on the first try. The pricing FAQ says purchased usage credits are what a Pay as you go account spends. This guide did not find a refund for unused credits.

Put it in

1. 1

   Step 1

   Install the local app

   The README shows a shell install on macOS and Linux, a Windows install script, and the Docker image ollama/ollama. The LICENSE file is MIT. This guide did not run the install. The README's first run is the ollama command, then a model name.

2. 2

   Step 2

   Pull one model you already named

   The README shows ollama run gemma4 and points at ollama.com/library. Run one name from your list. If the machine cannot hold it, pick a smaller library entry. The weights download onto that machine. Cloud concurrency, which the Pay as you go card lists as 3 requests, is not a local setting this guide found.

3. 3

   Step 3

   Point one client at port 11434

   The README shows a chat request to http://localhost:11434/api/chat with a model field and a messages array. Change one script to that URL and the local model name. Stop if the reply is empty. The README also links Python and JavaScript libraries. This guide did not find a page that maps a hosted request onto that local URL field by field.

What you lose

• The credit balance stays on the account

  Pro at $20 a month includes $60 of usage credits. Max at $100 includes $300. Team at $500 includes $1,000 shared. A local install does not read that balance. Unused included usage does not roll over, which the pricing FAQ says in those words.

• Hosted capacity is the thing you gave up

  A model that does not fit on your hardware still needs the cloud, and that call is billed per token. The pricing FAQ says prompts on the hosted service are not logged or trained on, and that capacity is primarily in the United States with possible routing to Europe. A local run does not use that host.

• Open WebUI is a screen, not the server

  The Ollama README lists Open WebUI as a web interface. It is not scored on this checklist. It does not replace the model download.