Open WebUI with Docker as an internal chat frontend

Open WebUI is a self-hosted chat interface that talks to local language models through Ollama or an OpenAI-compatible API. This tutorial starts it in Docker and connects it to a model server in your own network.

Niklas Schult4 min read

Illustration: An open wooden crate holding a cardboard speech bubble, linked by a hose to a tap, stands for an internal chat frontend.

What you will have at the end and what you need

At the end, a container named open-webui runs on one of your servers and serves a browser interface on port 3000. Colleagues open it like any ChatGPT-style page, while the model behind it is one you operate yourself. The project describes Open WebUI as a self-hosted platform built to run entirely offline, and it connects local models through Ollama or OpenAI-compatible APIs.

You need a host with Docker installed and a reachable model server, usually Ollama. A GPU is optional and only matters for the image variants covered later. The Open WebUI README also describes a pip install and a desktop app, but this article uses only Docker. Your IT team should decide whether your network design suits this setup, and the page on local AI gives background on running models in-house.

Step 1: Start the container

The following command starts the standard image in the background. It maps host port 3000 to port 8080 in the container and adds the name host.docker.internal so the container can reach services on the host. It also stores data in a volume called open-webui and restarts the container automatically.

Shell
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

Afterwards you open http://localhost:3000 on that machine. Colleagues use the server's address in your network with the same port.

Step 2: Connect Ollama or another server

If Ollama runs on a different machine, you pass its address through the environment variable OLLAMA_BASE_URL. The next command is the same start as before with that variable added, and you must replace the placeholder host name with your own server. Remove the first container before you start this one, since the name open-webui is already taken.

Shell
docker run -d -p 3000:8080 \
  -e OLLAMA_BASE_URL=http://your-ollama-server:11434 \
  -v open-webui:/app/backend/data \
  --name open-webui --restart always \
  ghcr.io/open-webui/open-webui:main

The README states that Open WebUI also works with OpenAI-compatible APIs for local and cloud models, and it lists an OPENAI_API_KEY option. It does not show how to point the container at a self-hosted OpenAI-compatible server, so the documentation leaves that step open. If the goal is to keep data inside your network, make sure the key you set never belongs to a cloud provider.

Choosing the image and the graphics card

The README lists several image tags. The main tag is the standard image with local AI dependencies, and slim is a lightweight image that relies on external services for embeddings, document extraction and voice. The cuda tag adds CUDA acceleration, the ollama tag bundles Open WebUI with Ollama, and dev is the latest development build.

The README says the gpus option enables Nvidia GPU access for the cuda or ollama images and needs the Nvidia Container Toolkit on the host. A second option, HF_HUB_OFFLINE=1, disables Hugging Face model downloads for offline use. Both options are added to docker run before the image name, so you can combine them with the commands shown above.

Which tag suits you depends on where the model runs. If Ollama already runs on a separate server, the main or slim image is enough. The bundled ollama tag only makes sense when one host should carry both the interface and the models.

Checking, troubleshooting and updating

The second prints its log, which is the first place to look when the page does not load or no models appear.

Shell
docker logs open-webui

For connection problems the README points to a troubleshooting guide, and for upgrades to an updating guide, both in the project documentation. The README excerpt gives no update commands, so follow that guide. Security vulnerabilities are reported privately through GitHub security advisories.

Where this fits in a company

A plain Open WebUI installation is a chat window. Someone in IT has to own the host, the model server, backups of the volume and updates. Settle questions about personal data in chats, a works council agreement (Betriebsrat) and duties under the GDPR before rollout. The licence requires preserving Open WebUI branding, which is worth checking before you customise the interface.

A chat frontend does not search your documents or cite sources on its own. souveraen.ai is a platform by clouditiv GmbH that answers from connected company sources with citations, and it runs hosted in Germany or on-premise. If you want to compare the two routes, see on-premise options and the model catalogue.

Sources

More articles

All articles

How safe are AI API routes via resellers and routers?

Security firm Dragos sent about two million requests with fake credentials to AI providers and found compromised services and resellers reading customer prompts. Companies that use AI through third parties need to know who sits on the path of their data.

Data protection4 min

Run Text Generation Inference with Docker on Your Server

Text Generation Inference (TGI) serves open language models from a single Docker container on your own hardware. This tutorial starts it, sends requests over HTTP and keeps prompts inside your network.

Tutorial4 min

Cloud outage liability: who answers when Microsoft fails

On 12 November 2026 a software fault took Exchange Online and Outlook offline for about 36 hours. A legal analysis published shortly afterwards argues that responsibility for such outages often lies with the operator rather than the provider.

Regulation4 min

Test it with your own documents.

Free for 3 days, then from €24.90 per user per month. Cancellable monthly.

Free for 3 days, then from €24.90 per user per month.Free for 3 days
Try free for 3 days