ai-expertise.eu

Open WebUI: Use AI Models on Your Own Terms

Verfasst von

juliluliluj

in

An Open-Source AI Interface You Can Run Yourself

Most people interact with AI through ChatGPT, Google Gemini, or Microsoft Copilot. These are solid products, but they come with trade-offs: your conversations pass through US-based servers, you’re limited to one provider’s models, and you have little control over how your data is handled. If that bothers you — or if you simply want more flexibility — there’s an open-source alternative worth knowing about.

It’s called Open WebUI.

What Is Open WebUI?

Open WebUI is a free, open-source AI chat interface that you install on your own computer or server. It connects to AI models via APIs — whether those models run locally on your machine or remotely at a provider like Mistral, OpenAI, or Google. The interface itself is self-hosted, meaning it runs under your control.

The project has over 50,000 stars on GitHub and more than 700 contributors. It’s actively maintained, well-documented, and looks polished enough that you wouldn’t guess it’s a community project. It’s licensed under a permissive open-source licence, so anyone can use, inspect, and modify the code.

What Does It Offer Beyond ChatGPT or Gemini?

If you already use ChatGPT or Gemini, you might wonder what the point is. Here are the practical differences:

Choose Your Own Provider

ChatGPT only gives you access to OpenAI’s models. Gemini only gives you Google’s. Open WebUI connects to any provider with an OpenAI-compatible API — and most providers offer one. You can configure multiple providers simultaneously and compare responses side by side. Switching from one provider to another is a matter of changing a URL and an API key in the settings.

More Control Over Your Data

Open WebUI itself runs on your machine, so your chat history, settings, uploaded documents, and user accounts are stored locally — not on someone else’s server. It’s worth being precise here: when you use a remote API (whether that’s OpenAI, Mistral, or anyone else), your prompts and the model’s responses do travel to that provider’s servers for processing. What changes is that the provider doesn’t also control the interface, your stored conversations, or your documents. And if you run a local model via Ollama, then nothing leaves your machine at all.

Document Chat (RAG)

You can upload PDFs, Word documents, or text files and ask questions about their content. Open WebUI uses Retrieval Augmented Generation (RAG) to pull relevant passages and shows inline citations so you can verify where an answer came from. It supports 9 different vector databases and multiple document extraction engines. This is a feature that commercial platforms offer in limited form, but Open WebUI’s implementation is unusually flexible.

Web Search Integration

Open WebUI can search the web and include results in your conversation context. It supports over 15 search providers, including privacy-oriented options like SearXNG and DuckDuckGo.

Image Generation

You can generate and edit images directly in the chat interface — using DALL-E, FLUX models via Black Forest Labs, or locally running engines like ComfyUI and AUTOMATIC1111.

Voice Interaction

The interface supports speech-to-text and text-to-speech with multiple engines, including fully local options via Whisper. If you use a local speech engine, the audio data stays on your machine.

Multi-User and Teams

You can create accounts for multiple users with role-based access control. There’s SSO/LDAP integration for organisations that need it.

Plugins and Extensibility

Open WebUI has a Pipelines framework that lets you add custom logic in Python — things like rate limiting, usage monitoring, translation, or content filtering.

You Don’t Need a GPU

A common misunderstanding: many people think self-hosted AI means you need an expensive GPU to run models locally. That’s not the case here. Open WebUI is the interface, not the model. The models run elsewhere — typically at an API provider.

In practice, this means:

  1. Install Open WebUI on any regular computer (it’s lightweight)
  2. Connect it to an API provider — for example Mistral, Scaleway, or if you already have one, your existing OpenAI or Google API key
  3. Use it — with all the extra features described above

Running local models with Ollama is an option for later, if you have a GPU with enough VRAM (8 GB or more). But it’s entirely optional.

Setting It Up

The installation is straightforward. There are two common approaches:

With Docker

  1. Install Docker Desktop (free, available for Windows, Mac, and Linux)
  2. Open a terminal and run:
    docker run -d -p 3000:8080 -v open-webui:/app/backend/data --name open-webui ghcr.io/open-webui/open-webui:main
  3. Open http://localhost:3000 in your browser
  4. Create an admin account

That’s one command after installing Docker. There’s also a Docker Desktop Extension that reduces this to a single click.

With Python

pip install open-webui && open-webui serve

Both methods take a few minutes. The official documentation covers additional options including Kubernetes and Podman.

API Providers Worth Considering

One of the practical advantages of Open WebUI is that it decouples the interface from the model provider. You can start with whatever API you already have access to, and add or switch providers at any time. Here’s an overview of providers whose APIs are compatible with Open WebUI:

Mistral AI 🇫🇷

Mistral AI is a French company that has become one of the most competitive AI providers globally. Their API is OpenAI-compatible, so it works with Open WebUI without any special configuration — just set the API URL and key.

  • Mistral Large 3 — flagship model, strong in reasoning and multilingual tasks
  • Mistral Medium 3.1 — good balance of capability and cost ($0.40/$2.00 per million tokens input/output)
  • Mistral Small 3.2 — fast and efficient for everyday use
  • Codestral — optimised for code generation
  • Voxtral — audio transcription

Many Mistral models are open-weight, meaning you can also download and run them locally. Their models are particularly strong in European languages.

Scaleway Generative APIs 🇫🇷

Scaleway, a French cloud provider, offers serverless API endpoints for popular open-source models — hosted in European data centres. The API is OpenAI-compatible. You get access to models from Mistral, Meta (Llama), and others, billed per million tokens. This is a good option if you want to use open-source models without running them yourself, with data staying in the EU.

OVHcloud AI Endpoints 🇫🇷

OVHcloud, Europe’s largest independent cloud provider, offers AI Endpoints — a serverless API for over 40 open-source models. Fully European infrastructure, OpenAI-compatible API, pay-per-use pricing.

NLP Cloud 🇫🇷

NLP Cloud is a French API provider that offers various open-source models on dedicated EU infrastructure, with a focus on privacy.

Black Forest Labs 🇩🇪

Black Forest Labs, based in Freiburg, builds the FLUX image generation models. Founded by the original creators of Stable Diffusion. Their API offers image generation starting at $0.014 per image. The FLUX.1 [schnell] model is Apache 2.0 licensed and can be run locally for free.

OpenAI, Google, and Others

If you already have an OpenAI or Google API key, those work too. Open WebUI doesn’t force you to choose one provider — you can configure several and use them in parallel. Services like OpenRouter aggregate many providers behind a single API, which can be convenient for trying different models.

Why Provider Diversity Matters

When you use ChatGPT or Gemini directly, you’re locked into one company’s ecosystem — their models, their interface, their terms, their content policies. If they change their pricing, restrict a feature, or alter how they handle your data, you have no recourse other than to stop using the service.

With Open WebUI, switching providers is a configuration change. This has a few practical implications:

  • No lock-in. Your chat history, documents, and settings live on your machine, independent of any provider. If a provider raises prices or shuts down, you change a URL and continue working.
  • Legal clarity. If you’re in Europe and care about GDPR compliance, you can choose providers that process data within the EU under EU law. The US CLOUD Act allows American authorities to compel US companies to hand over data regardless of where it’s stored — a legal conflict that European institutions are increasingly taking seriously. Using a European provider avoids this issue entirely.
  • Competition is healthy. A diverse market with multiple viable providers — American, European, and others — leads to better products, lower prices, and more innovation. When users can easily switch, providers have to compete on merit.
  • Different models have different strengths. Mistral models tend to handle European languages well. Some models are better at code, others at creative writing. Having access to multiple providers through one interface lets you use the right tool for each task.

What Open Source Means Here

Open WebUI’s source code is publicly available on GitHub. In practical terms:

  • Transparency. Anyone can inspect the code to verify there’s no hidden data collection or tracking.
  • Continuity. The software doesn’t depend on a single company’s survival. If the project were ever abandoned, the code would still be available and usable.
  • Customisation. If you need a specific feature or integration, you can build it — or find someone who has. The Pipelines plugin system is designed for this.
  • Local data storage. Your conversations, documents, and settings are stored in a local database. You can back it up, migrate it, or delete it at any time.

A Practical Starting Point

If you want to try this, here’s a reasonable approach:

  1. Install Open WebUI using Docker or Python (see above)
  2. Create an API account at Mistral or Scaleway (or use an existing OpenAI/Google key if you have one)
  3. Add the provider in Open WebUI’s settings (Admin → Connections → add the API URL and key)
  4. Start using it — try uploading a document, comparing models, or setting up a custom system prompt

Later, you can add more providers, experiment with local models via Ollama, or set up accounts for other people.

Links

This article is part of the ai-expertise.eu project, where we write about practical approaches to using AI independently. The article was written by AI with human assistance.

Weitere Beiträge