Ollama Compose Stack

Local Docker Compose setup for running Ollama with Open WebUI and SearXNG search support.

Services

• ollama: Ollama server exposed on http://localhost:11434
• open-webui: Open WebUI exposed on http://localhost/
• searxng: SearXNG server exposed on http://localhost:8080
• valkey: backing store for SearXNG

Open WebUI connects to Ollama through the Compose network at http://ollama:11434. Authentication is disabled for this local deployment. Hugging Face offline mode is enabled so the UI does not block startup on external model downloads. Open WebUI RAG embeddings are configured to use Ollama with nomic-embed-text. Open WebUI web search is enabled through SearXNG at http://searxng:8080/search?q=<query>, returning 5 results with 2 concurrent requests.

HTTPS is expected to terminate on the separate Caddy host. Point that Caddy reverse proxy at http://<this-host>:80 and keep WebSocket proxying enabled.

Setup

Create a local .env file from the example:

cp .env.example .env

Then replace the placeholder values:

SEARXNG_SECRET=change-this-to-a-long-random-string
OLLAMA_API_KEY=change-this-to-a-long-random-string

The .env file is ignored by git and should not be committed.

Run

Start the stack:

docker compose up -d

Check the resolved configuration:

docker compose config

Stop the stack:

docker compose down

Pull Models

Pull models from the included list:

./pull-model.sh models.list

Pull a single model:

./pull-model.sh qwen3:8b

Pull multiple models:

./pull-model.sh qwen3:8b,mistral:7b

Model list files support blank lines and # comments.

Local Data

Ollama data is stored in ollama-data/. This directory can contain model blobs, history, and local SSH keys generated by Ollama, so it is ignored by git.

Open WebUI data is stored in open-webui-data/. This directory can contain chat history, users, settings, and other local application state, so it is also ignored by git.