0Orientation & objectives
This course covers the Open WebUI + Ollama stack as a whole: what each half does, how Docker packages and runs both, how to wire them together, where configuration and security defaults live, how to use GPU acceleration and quantization to keep it fast, and the maintenance habits that prevent the most common support problems. Audience: a self-hoster setting up this stack for the first time, comfortable with a terminal but new to this specific combination.
| Element | Detail |
|---|---|
| Format | Self-paced; approximately 35–40 minutes |
| Formative questions | 16 multiple-choice, distributed (feedback, not scored) |
| Final assessment | 14 multiple-choice questions; pass mark 75% |
| Sources | This repository's wiki (official documentation, 2026-07-07/08) |
1The Stack, Overview
Open WebUI is an extensible, self-hosted AI platform designed to run entirely offline, working with both Ollama and OpenAI-compatible APIs rather than locking users into one model provider.
Ollama is local large-language-model infrastructure built on llama.cpp. Its own scope stays narrow — model serving — while a large third-party ecosystem, including Open WebUI, builds interfaces and tooling around it.
The two connect asymmetrically. Open WebUI is deeply aware of Ollama specifically: a dedicated
configuration variable, OLLAMA_BASE_URL (default http://localhost:11434, different
inside Docker networking), points at a running Ollama server, and Open WebUI's backend proxies straight
through to Ollama's native API. There is no reimplementation of inference logic on the Open WebUI side —
Ollama is just another connected model provider. Ollama, for its part, is not aware of Open WebUI at all; it
lists Open WebUI only as one more community front end in its own documentation.
Check your understanding
OLLAMA_BASE_URL variable do?