Ollama: Open Web UI

This is a guide on how to integrate ollama with Open WebUI for self-hosting LLMs.

First install Docker. See the Docker page for local setup.

Quick Start

If you only want to launch Open WebUI itself and connect it to an existing Ollama instance, the official quick-start is:

docker run -d -p 3000:8080 \
  --add-host=host.docker.internal:host-gateway \
  -v open-webui:/app/backend/data \
  --name open-webui \
  --restart always \
  ghcr.io/open-webui/open-webui:main

Then open http://localhost:3000.

Open WebUI automatically attempts to connect to an Ollama instance. If that fails, check the Ollama connection settings from the admin panel after the first login.

NVIDIA GPUs

The official NVIDIA-enabled image uses the :cuda tag and Docker GPU support:

docker run -d -p 3000:8080 \
  --gpus all \
  -v open-webui:/app/backend/data \
  --name open-webui \
  --restart always \
  ghcr.io/open-webui/open-webui:cuda

This only handles the Open WebUI container itself. Your Ollama setup still needs to expose a working model backend separately.

AMD GPUs

sudo pacman -S docker docker-compose

Take for reference this docker compose file that was posted on reddit

services:

  ollama:
    image: ollama/ollama:rocm
    container_name: ollama
    environment:
      OLLAMA_MODELS: /usr/share/ollama
      HSA_OVERRIDE_GFX_VERSION: "X.Y.Z"
      HIP_VISIBLE_DEVICES: "0"
    devices:
      - "/dev/kfd"
      - "/dev/dri"
    security_opt:
      - seccomp:unconfined
    cap_add:
      - SYS_PTRACE
    ipc: host
    group_add:
      - video
    volumes:
      - /home/username/.ollama:/root/.ollama
      - /home/username/ollama/models:/usr/share/ollama
    ports:
      - "11434:11434"
    networks:
      backend:
        ipv4_address: 10.1.0.2

  open-webui:
    image: ghcr.io/open-webui/open-webui:cuda
    container_name: open-webui
    environment:
      OLLAMA_BASE_URL: http://10.1.0.2:11434
      HSA_OVERRIDE_GFX_VERSION: "X.Y.Z"
      HIP_VISIBLE_DEVICES: "0"
    devices:
      - "/dev/kfd"
      - "/dev/dri"
    security_opt:
      - seccomp:unconfined
    cap_add:
      - SYS_PTRACE
    ipc: host
    group_add:
      - video
    volumes:
      - /home/username/.open-webui:/app/backend/data
    ports:
      - "3000:8080"
    networks:
      backend:
        ipv4_address: 10.1.0.3

networks:
  backend:
    driver: bridge
    ipam:
      config:
        - subnet: 10.1.0.0/16
          gateway: 10.1.0.1

Create in a project directory a docker-compose.yml, and execute

docker compose up -d

Don’t forget to substitute the correct username and X.Y.Z GFX version in the compose file. See the Ollama Guide for more information.

Then access http://localhost:3000 in your web browser. You should be able to see the Open WebUI login screen.

To stop the services

docker compose down

There might be conflicts with the default port for ollama.service, run
sudo lsof -i :11434
to verify. Either stop the ollama daemon, or change the docker compose file to another port.