Local LLMs: Running Ollama and Open WebUI in Docker on Ubuntu

The post describes a setup where Ollama runs as a local service on Ubuntu and Open WebUI runs in a Docker container. It explains why the author prefers this split, with Ollama using the machine directly and Open WebUI staying available in the background.

Reading notes

#

• Docker is required, and the author mentions Docker Desktop or Docker Engine as installation options.
• Ollama is installed on Ubuntu through the Linux installation instructions and runs as a local service.
• By default, Ollama listens on 127.0.0.1, so the post changes the service configuration to listen on 0.0.0.0:11434.
• The Ollama service override sets OLLAMA_HOST=0.0.0.0:11434 and OLLAMA_CONTEXT_LENGTH=32768.
• The post notes that exposing Ollama on 0.0.0.0 makes it reachable from the local network and suggests firewall restrictions for security.
• A model is pulled with ollama pull gpt-oss:20b, and other models such as llama3, mistral, or gemma are mentioned as options.
• Open WebUI is started with docker run, mapping port 9090 to 8080, naming the container open-webui, and mounting $HOME/.open-webui for persistent data.
• The container uses –add-host=host.docker.internal:host-gateway so it can reach services on the host.
• After startup, the post says to wait for the container status to show Up and (healthy), then open http://localhost:9090.
• In Open WebUI, the user selects the pulled Ollama model from the model dropdown and starts chatting.
• When a new Open WebUI version is available, the post says to stop the container and rerun the docker command with the new image tag.
• An alternative setup uses –network=host and OLLAMA_BASE_URL=http://127.0.0.1:11434/ instead of changing Ollama’s host binding.
• The author prefers the bridge-network approach because it keeps the container limited to services that are explicitly reachable on the host network.
• Cleanup commands include docker rm open-webui –force, sudo systemctl stop ollama, and sudo systemctl disable ollama.