Open WebUI for beginners: install with Ollama

By Emmanuel Sauvage · Published · Updated

This page may contain Amazon affiliate links. A purchase after a click can earn the site a commission, at no extra cost to you. The advice stays independent. How affiliation works.

Docker alone: plan on 45 minutes, plus 20 minutes if you use the pip appendix (the home network is covered in the note after the first chat).

Open WebUI is a friendly web page that connects to Ollama (the engine that runs LLMs, large language models, on your PC). Each person in the household opens a page in the browser instead of typing in a terminal. This guide follows one thread: Docker + Ollama, on the time above. It assumes Ollama is already installed (Windows, every OS, or the 15-minute path).

Who Path Time What you need
Beginner, Docker already installed Sections 1 to 4 below 45 min (Docker only) Ollama running, 16 GB RAM
No Docker Pip appendix after the comparison table +20 min (about 65 min total) Python 3.10+
Family on tablets The network note after the first successful chat +10 min on top Docker or pip already working

On this page

What you need

Piece Detail
Ollama Service running, default port 11434 — install Ollama if you still need to
RAM 16 GB minimum, 32 GB recommended for a 7B–8B model (32/64 GB of RAM)
Rights An administrator account to install Docker or Python
Network Internet to download Open WebUI once. Chat can work offline after that
Model At least one model already downloaded (choose a first model)
Disk Room for Docker images or a virtual environment (SSD for models if you are stacking GGUF files)

Docker or pip: a comparison

Criterion Docker pip / native binary
Simple when Docker is already there ★★★ ★★
What you need first Docker Desktop or a Linux engine Python 3.10+ or the official package
Updates docker pull (see the update appendix) A command or an installer, depending on the docs
Network isolation Can cause CORS errors (Cross-Origin Resource Sharing, the rules for how a browser may call an API) if the setup is off Often simpler when everything stays on one machine
A household A stack close to Docker Ollama + Open WebUI Fewer containers

How to read the stars: ★ means a long path or many dependencies; ★★ means a few extra commands; ★★★ means few steps once the base tool is installed.

If Docker is not installed, open the pip appendix. Run one method on the machine, and stop the other service before you switch.

Install steps (Docker — the path to use when Docker is installed)

  1. Check Ollama. In a terminal:
curl http://127.0.0.1:11434/api/tags

You should get JSON (a list of models), or an empty reply {"models":[]} if no model has been downloaded yet.

  1. Download a model if the list is empty:
ollama pull llama3.2:3b
  1. Start Open WebUI. Copy and paste the full command as it is:

Copy the whole block. On Windows PowerShell, use one line with backticks ` at the end of each continued line, or paste the multi-line block as it is into cmd, Linux, or macOS.

docker run -d \
  -p 8080:8080 \
  --add-host=host.docker.internal:host-gateway \
  -v open-webui:/app/backend/data \
  --name open-webui \
  ghcr.io/open-webui/open-webui:main

Optional breakdown (the same command, so you can see each piece. Skip this if the block above already works):

Piece What it does
-d Container in the background
-p 8080:8080 Browser port mapped to the container
--add-host=host.docker.internal:host-gateway Reach Ollama on the host from inside Docker
-v open-webui:... Keep accounts and history
ghcr.io/... Official image (check the docs if the tag changes)
  1. Open the browser at http://localhost:8080 (the local address of your machine).

  2. Create the administrator account. The first account you register becomes admin (the person who manages users and settings).

  3. Choose the default model in settings. Select llama3.2:3b, or the model you downloaded.

  4. Send a test message: “Summarize in three sentences why local AI can help at home.”

A reply in English means the install is confirmed.

After your first successful chat

Network, household, and data (one note) — While you are testing alone on the PC, stay on localhost. For a tablet on the same Wi-Fi, open http://YOUR-PC-IP:8080 (the LAN, the local Ethernet or Wi-Fi network). Find the IP with ipconfig (Windows), ip a (Linux), or Network settings (macOS). Leave ports 8080 and 11434 closed on the router toward the internet until you have a VPN (a virtual private network, an encrypted tunnel) and a strong login. The first account you register stays admin: pick a strong password. Before attachments or someone else’s data, read fully local AI and privacy. For accounts and multi-user habits, see the private family assistant. Back up the Docker volume open-webui if the history matters — see SSD storage for models.

Troubleshooting

Port 8080 is taken, or the container is missing

Symptom: the container does not start, or the browser shows a different site.

Fix What to do
Change the port Use -p 8081:8080 in docker run, then open http://localhost:8081
Free 8080 docker ps, find the container, then docker stop name
Check the firewall Allow the port you chose on the private network only (see the network note above)

Symptoms and fixes

Symptom Likely cause Fix
Model list is empty No ollama pull yet ollama pull llama3.2:3b, then refresh the page
502 / backend Ollama is stopped Start Ollama again; curl http://127.0.0.1:11434/api/tags
No model (Docker) Ollama URL or CORS (Cross-Origin Resource Sharing, the browser-to-API rules) Set OLLAMA_BASE_URL=http://host.docker.internal:11434 (Docker Desktop) or the Linux host IP, then restart the container
Blank page Container stopped docker ps; docker start open-webui
Port 8080 taken Another service Change the port or free 8080 (see above)
Slow interface Model is too heavy A 3B or a Q4 — choose a first model
Slow on the LAN Wi-Fi is saturated Plug the server PC into Ethernet, and reread the network note
Forgotten password Local account The reset steps in the Open WebUI docs, or reinstall while keeping the volume
PC left on around the clock Electricity use kWh budget

What a finished install looks like

  • A local web page with chat history.
  • An admin account created, and a default model selected.
  • An English reply in under a minute on a recent PC.

Mistakes that show up before anyone buys hardware

Mistake Why it stalls Fix
Start Open WebUI with Ollama stopped Empty backend, 502 errors curl http://127.0.0.1:11434/api/tags, then start Ollama
A 70B model on 16 GB RAM The disk swaps, and the PC locks up Go back to llama3.2:3b or a Q4 7B — choose a first model
LM Studio, Ollama, and a GPU game at once VRAM is full Close the other GPU apps. LM Studio stays optional
History gone after docker pull A new container with no volume Keep -v open-webui:/app/backend/data on every docker run

Open WebUI shares the same PC as Ollama. You can finish this tutorial on the machine you have. If you have already lightened the model (3B, then 7B) and the wait still bothers you day to day, read VRAM and models and the hardware selector to frame RAM and GPU before a purchase. To compare 32 GB mini PCs on the spec sheet, a search for “minisforum 32gb” on Amazon is a starting point. Affiliate link: a commission is possible at no extra cost to you. Check the current price and whether it is in stock on the store page (how affiliation works, October 2026). The mini PC comparison is the neutral read when you want to decide.

Appendix — pip install (no Docker)

If Docker is not installed, the Open WebUI docs often recommend pip (Python’s package manager):

  1. Check Python 3.10 or newer:
python3 --version
  1. Create a virtual environment (recommended), then install the official package. Confirm the version on docs.openwebui.com when you install:
python3 -m venv ~/.open-webui-venv
source ~/.open-webui-venv/bin/activate
pip install open-webui
  1. Start the server pointed at local Ollama:
export OLLAMA_BASE_URL=http://127.0.0.1:11434
open-webui serve
  1. Open the URL printed in the terminal (often http://localhost:8080), create the admin account, and pick a model you already downloaded with ollama pull.

If dependencies fail, go back to the Docker/pip table. A native binary with no pip: manual install (official docs) or the “Desktop” package on docs.openwebui.com. The sequence is the same (web server, then admin account, then the Ollama backend at http://127.0.0.1:11434).

Appendix — update Open WebUI (Docker)

Back up the open-webui volume if the history matters, then:

docker stop open-webui
docker rm open-webui
docker pull ghcr.io/open-webui/open-webui:main
docker run -d \
  -p 8080:8080 \
  --add-host=host.docker.internal:host-gateway \
  -v open-webui:/app/backend/data \
  --name open-webui \
  ghcr.io/open-webui/open-webui:main

Match the command against docs.openwebui.com if the image tag changes. A documented multi-container stack: Docker Ollama + Open WebUI.

Open WebUI questions (for this guide)

Why does the browser show a CORS error with Docker?

CORS (Cross-Origin Resource Sharing, the rules between the browser and the Ollama API) sometimes blocks Docker when the backend URL is wrong. Set OLLAMA_BASE_URL=http://host.docker.internal:11434 (Docker Desktop), or the host PC’s LAN IP on Linux, then restart the container. Detail is in the troubleshooting table.

Can LM Studio stand in for Open WebUI for the family?

LM Studio gives you a graphical screen for trying GGUF files. Ollama is still the usual API backend for Home Assistant or Continue. For a household that mostly chats in the browser, Open WebUI plus Ollama is enough. See LM Studio GPU offload alongside it, and keep both off the same GPU at once.

Do the conversations go out to the internet?

They stay on the machine while ports 8080 and 11434 stay closed to the internet and you have not turned on a cloud sync. LAN, admin, backup, and the legal pointer live in the network and data note. Compare local cost vs ChatGPT if you are choosing between hardware and a subscription.

Once localhost:8080 is stable, continue without stacking too many services:

  1. Private family assistant — accounts and multi-user use on the same Open WebUI instance (it fills out the network note).
  2. Docker Ollama + Open WebUI — a stack you can repeat (several PCs, or identical backups).
  3. First steps with Whisper — offline speech-to-text after text chat works.
  4. An SSD for models — Docker volumes and ollama pull folders before the GGUF files pile up.

Time the same question on a 3B, then on a 7B. If the wait is still acceptable, put off a dedicated mini PC and adjust 32 vs 64 GB of RAM first.

Common questions

Docker, or a Python (pip) install?

See the Docker versus pip comparison table in this guide. In short: Docker if it is already installed, otherwise pip or a native binary (the collapsible appendix at the bottom of the page).

Can I open Open WebUI from a phone?

Yes, on the local network (LAN), if the firewall allows the port and you leave the service off the public internet until the login is strong. See the network note after the Docker install.

The model is missing from the list. What now?

Run ollama pull and the model name in a terminal, then refresh Open WebUI. See the Troubleshooting table.

Does Open WebUI replace Ollama?

Open WebUI is the web page. Ollama stays the engine that runs the LLMs in the background.

Sources

Narrow it down with the hardware selector · French version of this section