How to run Ollama with Docker

The official Ollama Docker image runs the server in a container while storing model data in a named Docker volume. That keeps the local host clean and gives automation a repeatable way to start the API on port 11434.

CPU-only Docker usage needs only the image, a persistent volume, and a port mapping. GPU usage needs additional device or runtime configuration and belongs in a separate GPU-focused flow.

Use a named volume from the first run. Removing a container without a volume can also remove downloaded models, which makes later smoke tests appear slower or broken.

1. Create or reuse a named Docker volume for models.

   $ docker volume create ollama
   ollama

2. Start the official Ollama container on port 11434.

   $ docker run -d \
     -v ollama:/root/.ollama \
     -p 11434:11434 \
     --name ollama \
     ollama/ollama
   f3b2d2f3b0f1

3. Check that the container is running.

   $ docker ps --filter name=ollama
   CONTAINER ID   IMAGE           COMMAND          CREATED          STATUS          PORTS                      NAMES
   f3b2d2f3b0f1   ollama/ollama   "/bin/ollama"   10 seconds ago   Up 10 seconds   0.0.0.0:11434->11434/tcp   ollama

4. Verify the API through the host port.

   $ curl -s http://localhost:11434/api/version
   {"version":"0.31.1"}

5. Run a model inside the container after pulling it.

   $ docker exec -it ollama ollama run gpt-oss:20b "Return OK."
   OK

   Large model pulls can take significant time and disk space.