oow_ctl

A lightweight Bash wrapper for running Ollama and Open WebUI with Docker Compose on a GPU-equipped Linux host.

Overview

oow_ctl manages two Docker containers:

• Ollama — serves large language models locally, with full NVIDIA GPU passthrough.
• Open WebUI — provides a browser-based chat interface connected to Ollama.

Both services store persistent data on an external volume (/mnt/usb/docker/) via bind mounts.

Prerequisites

• Docker Engine with the Compose plugin
• NVIDIA Container Toolkit for GPU support
• A mounted volume at /mnt/usb/docker/ (or adjust the paths in oow_ctl.sh)

Note: This script is designed for a system with limited storage on the primary SSD where / is mounted (Asus Ascent GX10, similar to an NVIDIA DGX Spark). Additional space is available on a storage device mounted at /mnt/usb/, which is why Docker volumes are stored there instead of on the root filesystem.

Quick start

chmod +x oow_ctl.sh
./oow_ctl.sh start

Open WebUI will be available at http://localhost:8080.

Usage

./oow_ctl.sh {df|logs|restart|start|status|stop|top}

Command	Description
df	Show free disk space on / and the data volumes
logs	Tail container logs (additional arguments are forwarded to docker compose logs)
restart	Stop and then start both containers
start	Start Ollama and Open WebUI in detached mode
status	Show running container status
stop	Stop and remove both containers
top	Show running processes inside containers

Configuration

All settings are defined as environment variables at the top of oow_ctl.sh:

Variable	Default	Description
OLLAMA_IMAGE_NAME	ollama/ollama	Ollama Docker image name
OLLAMA_IMAGE_TAG	0.20.0	Ollama Docker image tag
OLLAMA_PORT	11434	Host port for the Ollama API
OLLAMA_VOLUME	/mnt/usb/docker/ollama	Host path for Ollama data
OPENWEBUI_IMAGE_NAME	openwebui/open-webui	Open WebUI Docker image name
OPENWEBUI_IMAGE_TAG	latest	Open WebUI Docker image tag
OPENWEBUI_PORT	8080	Host port for Open WebUI
OPENWEBUI_VOLUME	/mnt/usb/docker/open-webui	Host path for Open WebUI data

License

This project is licensed under the MIT License. See the LICENSE file for details.