Use Open WebUI with open-source models hosted in the EU

Open WebUI is an open-source web interface for language models that you run yourself, usually with Docker or Kubernetes. It offers a chat experience similar to commercial assistants, plus user management with roles and groups, document upload and knowledge bases, web search, tools, prompts and model presets that admins can share across the organisation.

It connects to models through OpenAI-compatible connections. Any number of connections can be added, each with its own URL, key and model list. inference.eu works as one of these connections, so models appear in the model picker for the users and groups you allow.

In Settings > Admin > Connections, add a connection with the inference.eu base URL ending in /v1 and your API key. Open WebUI calls /models to discover the available models; you can restrict the list with Model IDs. Each connection can be enabled or disabled without losing its settings.

For automated deployments, set OPENAI_API_BASE_URLS and OPENAI_API_KEYS as environment variables, separating multiple connections with semicolons in matching order. Open WebUI stores these values in its database on first start; after that, changes made in the admin panel take precedence over the environment unless you disable persistent configuration.

For general chat, writing and summarising, gpt-oss-120b gives quick answers that keep the interface responsive. For document chat over long files and questions that need careful reasoning, glm-5.3 and kimi-k3 are strong choices. Teams that ask coding questions benefit from qwen3-coder-480b-a35b-instruct.

Admins can create workspace models on top of these, with a system prompt, attached knowledge and tools, and share them with specific groups. Set the function calling mode to native for models with dependable tool calling, so tools run through the model's own tool calls rather than prompt-based detection.

If Verify Connection fails with 401, the key is wrong or has a trailing space. A 404 usually means the URL is missing /v1 or has an extra path. If no models appear, check the Model IDs filter and that the connection is enabled.

If a change to OPENAI_API_BASE_URLS has no effect after a restart, the value saved in the database is being used; update the connection in the admin panel instead. If Docker cannot reach the URL, check the container's outbound network and proxy settings. If tools are not called, switch the model's function calling mode or choose a model with stronger tool support.

With inference.eu as the connection, every chat message, uploaded document excerpt and tool result sent to the model is processed in EU data centres and never used for training. We sign a data processing agreement under Article 28 GDPR, publish our subprocessor list and bill in euros.

Open WebUI stores chats, files and users in its own database, so host it in the EU. Web search, image generation, speech and embedding settings can each use a separate provider; configure EU-hosted or local options for them, and disable or remove connections you do not want users to reach.