Open WebUI
Want ChatGPT-style chat without handing every conversation to one company? Open WebUI gives you a friendly chat window you run yourself, connected to a model on your computer or a service you choose.
How you use it: Install the desktop app, connect it to a model on your computer or an AI service, and chat in a familiar window.
In plain English
The GitHub description calls Open WebUI a user-friendly AI interface that supports Ollama and the OpenAI API. Instead of living only in a terminal, it gives you a page in the browser, and the README also links a desktop app for macOS, Windows, and Linux.
You can install it with Python or with Docker. The Python path serves a page at http://localhost:8080. The Docker command in the README maps that same app to http://localhost:3000. Docker image tags in the README include a standard image, a lighter image, a CUDA image for Nvidia cards, an image bundled with Ollama, and a development build.
The README shows how to point the app at an Ollama server on another machine, how to set an OpenAI API key, and how to disable Hugging Face downloads for offline use. Its topics include RAG and MCP, but this README spends its words on install and connection options rather than on a feature tour.
Practical ways to use it
Example 1
Give a local model a chat window
Install the desktop app linked from the README, or run the Docker image tagged :ollama, and type to a model in the browser instead of a terminal.
Example 2
Use an OpenAI key without writing code
The Docker options table shows an OPENAI_API_KEY setting so the interface can call OpenAI.
Example 3
Reach an Ollama server on another computer
The README lists OLLAMA_BASE_URL so you can point the app at an Ollama server you already run.
Example 4
Work without downloading models from Hugging Face
The README shows HF_HUB_OFFLINE=1 to disable those downloads.
Who it is for
Non-coder. The desktop app is the path for someone who does not want a terminal. The Python and Docker installs are there if you like to run servers.
Difficulty: Easier start. The README links a native desktop app, and also documents Python and Docker for people who want the server.
Getting started
- 1
Start with the desktop app if you want the gentlest path
The README links a native desktop app for macOS, Windows, and Linux at github.com/open-webui/desktop.
- 2
Or install with Python
With Python 3.12, the README shows pip install open-webui and then open-webui serve. Open http://localhost:8080.
- 3
Or run the Docker image
The README's docker run command publishes the app on port 3000. Open http://localhost:3000. Use the :ollama tag if you want Ollama bundled.
- 4
Add a key only if you are using a hosted model
Set OPENAI_API_KEY when you want the OpenAI API. Skip it if you are only talking to a local Ollama server.
What you need first
- Accounts
- None for a local model through Ollama. An OpenAI API key is only for that provider, and the README shows where to pass it.
- Hardware
- The README does not give a minimum computer. The CUDA tag and GPU flag are for Nvidia cards, and they need the Nvidia Container Toolkit.
- Cost
- The app is what you install. A hosted model such as OpenAI charges for use. A local model costs you electricity and disk space. The README does not list prices.
Cautions
- The license note in the README says the Open WebUI License requires preserving the Open WebUI branding. Read LICENSE and LICENSE_HISTORY before you rebrand it.
- An API key in a Docker command can leak if you share the command or a screenshot. Treat it like a password.
- The project topics mention RAG and MCP, but do not assume a feature is present until you see it in the app or the docs. This README is mostly an install guide.
Questions people ask
What does Open WebUI do?
It gives you a user-friendly screen for AI chat, and it can connect to Ollama or to an OpenAI API key. A desktop app is linked from the README.
Who is Open WebUI for?
People who want a chat window rather than a terminal. The desktop app is the closest fit. Python and Docker are there if you prefer to run the server yourself.
What do I need first?
For the desktop app, the download. For the server, Python 3.12 or Docker. An OpenAI key is optional.
What should I watch out for?
Keep the branding if you share a copy, because the README says the license requires it. Do not publish your API key.