Open WebUI
by Open Source
Getting-started guide
The chat interface you run yourself — you bring the model, it brings everything else.
Current prices: Free · See all plans
Last reviewed 2026-09-07 · for Open WebUI
You install Open WebUI on your own computer and open it in a browser tab — except the site runs on your machine, not someone else's. That's what self-hosted means. What you type goes to whichever AI you've connected.
A browser-based chat window for AI you host yourself. It has no model of its own — the model is the AI "brain" that does the thinking — so you connect one separately.
Why this over Jan or LM Studio: those are desktop apps, one person, one computer. This one keeps running in the background, so your household reaches it from their own phones and logins.
Free, and the code is public — no account, no trial, no card. The company sells licensing to large organisations; nothing you'd run is paywalled.
Open WebUI is not an AI. Install it and you get an empty model dropdown — sort out the AI half first. Ollama is the usual answer, and this site has a page for it.
The real question isn't whether you're technical, it's whether you want a small server in your house. You'll need Docker (a way of running an app in a self-contained box, so you don't install its parts by hand). If you only want AI on your laptop, Jan or LM Studio do it in an afternoon.
Official source: https://openwebui.com/
Install only from the official project — the site above, or open-webui/open-webui on GitHub. Lookalikes exist, and this one holds your keys and chat history.
Follow its official quick-start, not a forum post; commands change between versions. Docker is the recommended path, Python and a desktop app the alternatives.
Open WebUI currently finds a local Ollama by itself in the standard Docker setup; an online provider needs its address and key in Settings.
Make the admin account yourself, immediately — the first account on a fresh install becomes the administrator and sign-up closes behind it. Then open the address the docs give you (currently port 3000 for Docker, 8080 for Python and desktop) and send one message; an empty model dropdown means no AI is connected.
Its privacy posture isn't fixed — your model source decides it.
Securing it is your job: keep it on your home network unless you have a reason not to. An instance left open is a stranger's free chat window on your hardware.
Points Open WebUI at a single model source — a local Ollama it finds by itself, or one online provider's address and key in Settings — and sends one message to confirm the dropdown fills.
Why this one: The settings surface is large, and the temptation is to configure everything before it works. Later problems are easier to diagnose from one known-good path.
Keeps the instance reachable only on your home network, makes the admin account yourself, and checks that sign-up closed behind it.
Why this one: Unlike a desktop app, this is a server the moment it starts — the only tool in this category where "who else can reach this?" is a question you answer yourself.
Explain in plain language why an AI on my own laptop might answer more slowly than one on a company's servers. Under 200 words.
Why this one: One question against a model on your machine and one online, in the same window, shows what this is for, and where a small local model's limits are.
Want the full picture first? See the Open WebUI overview for pricing, the free tier and related tools.
