Open WebUI with Ollama.
Private AI chat on your server.

Memory sizing

The container is light at idle but memory-hungry once a model is loaded, because the model weights are held in RAM by Ollama. Plan for at least 8 GB of RAM as a floor, and more for larger models. The catalog sets an 8 GB minimum for this reason. Choose model sizes that fit the RAM on your plan.

docker-compose walkthrough

The image ships a standalone docker-compose.yml:

name: open-webui

services:
  open-webui:
    image: ${OWUI_IMAGE:-ghcr.io/remarkablecloud/open-webui:0.11.3-r1}
    restart: unless-stopped
    ports: ["${HTTP_PORT:-8084}:8080"]
    environment:
      WEBUI_URL: ${WEBUI_URL:-http://localhost:8084}
    volumes:
      - owui-data:/app/backend/data
      - ollama-data:/root/.ollama

volumes:
  owui-data:
  ollama-data:

Line by line:

• image pins the RemarkableCloud Open WebUI tag, overridable with OWUI_IMAGE.
• restart: unless-stopped brings the container back after a crash or reboot.
• ports maps host port 8084 to container port 8080. This is for standalone runs only. Behind Traefik, no port is published; the proxy routes to port 8080 on the internal network.
• WEBUI_URL is the public base URL used for links.
• volumes mount two named volumes: owui-data at /app/backend/data (users, chats, settings in SQLite) and ollama-data at /root/.ollama (downloaded model weights).

Start it with docker compose up -d, open http://localhost:8084, and create the admin account. Then pull a model from Settings > Models, or run docker exec <container> ollama pull llama3.2.

Environment variable reference

Variable	App Platform value	Standalone default	Notes
WEBUI_URL	https://your-hostname	http://localhost:8084	Public base URL of the interface, no trailing slash. The platform fills the hostname.
HTTP_PORT	not applicable	8084	Host port mapped to container port 8080 in standalone compose only (not published behind Traefik).
OWUI_IMAGE	not applicable	ghcr.io/remarkablecloud/open-webui:0.11.3-r1	Image reference override for standalone compose.

Models are not configured with environment variables. Pull them from the UI (Settings > Models) after logging in; they are stored in the ollama-data volume.

Hardening notes

• First account is the admin. The first account created on a fresh install becomes the administrator. Create it right after deploy, then disable open sign-ups in the admin settings so strangers cannot register.
• TLS at the proxy. In production the container serves plain HTTP on 8080 and Traefik terminates TLS. Keep the container off any public port and let the proxy reach it internally.
• Embedded runtime stays internal. The bundled Ollama runtime is reachable only inside the container; this compose file does not expose it as a separate service.
• Runs as root. The upstream Ollama bundle runs as root. Keep it isolated behind the proxy and do not publish its port.
• Right-size the model. Loading a model larger than available RAM will fail or thrash. Match the model to your plan.

Backup and upgrade

Backup. Two volumes hold state. owui-data at /app/backend/data is the important one: it holds the SQLite database with users, chats, prompts, and settings. Back it up on a schedule. ollama-data at /root/.ollama holds downloaded model weights, which are large and can be re-pulled at any time, so you can exclude it from backups or place it on a lower tier to save space. For a consistent copy of the app data, snapshot the volume or stop the container briefly.

Upgrade. Pull the newer RemarkableCloud tag and recreate the container. Both named volumes persist, so your users, chats, and already-downloaded models carry over. Because state is SQLite, an upgrade is a container swap with no separate database migration step. Read the CHANGELOG for the target tag first, and back up owui-data before upgrading.