Open WebUI is a self-hosted web interface, close to ChatGPT, that connects to Ollama or to any OpenAI-compatible API. Docker installation, connecting Ollama, user accounts, documents, sizing and securing it on a GPU server.
By Mouhamed BANKOLE, CEO at ITNET Technologies · Updated November 6, 2026 · Reading time: 8 min
Your teams already use an online AI assistant, sometimes with internal documents, and you would rather keep that data in-house. Open WebUI lets you deploy your own chat interface, very close to ChatGPT, connected to models you host yourself. Combined with Ollama on a GPU server, it gives you an internal assistant with user accounts and history, with no per-person subscription. Here is what Open WebUI is, how to install it, connect it to Ollama, size it and secure it.
In short
- Open WebUI is a self-hosted AI platform that offers a ChatGPT-style chat interface and runs entirely offline.
- It connects to Ollama for local models, or to any OpenAI-compatible API.
- The simplest install uses Docker: one command, then the interface answers on port 3000.
- Size the server on the VRAM of the GPU that runs the model; the interface itself is lightweight.
- In production you need HTTPS, a secret key, named accounts and an Ollama instance that listens only locally.
Contents
- What is Open WebUI?
- Why host your own internal ChatGPT?
- How do you install Open WebUI?
- How do you connect Open WebUI to Ollama?
- Can you use Open WebUI with an OpenAI-compatible API?
- How do you manage users and documents?
- What server do you need for Open WebUI?
- How do you secure Open WebUI in production?
- How much does a self-hosted internal assistant cost?
What is Open WebUI?
Open WebUI is a self-hosted web interface for chatting with AI models, designed to run entirely offline and compatible with Ollama and OpenAI-compatible APIs. That is how the official Open WebUI documentation describes the project.
What it brings:
- a familiar interface: chat threads, history, model picker;
- several model providers: Ollama locally, or a remote OpenAI-compatible API;
- advanced features: search across documents (RAG), tools and extensions;
- deployment on your side: on a server, a VPS or a workstation.
The engine that actually runs the model is separate: it is often Ollama, which we cover in a dedicated guide.
Why host your own internal ChatGPT?
To keep control of the data sent to the AI, control costs and choose the models, without depending on a vendor. That is the main reason given by small businesses that handle sensitive documents.
- Privacy: questions and documents stay on a server you control, ideally in France or Europe.
- Predictable cost: one server per month rather than a per-user subscription.
- Model choice: Mistral, Llama, Gemma or a specialized model, depending on the use.
- Integration: a single access point for the whole team, with named accounts.
Our guide to hosting an AI application compares this approach with the other possible architectures.
How do you install Open WebUI?
The recommended method is Docker: a single command pulls the image, keeps data in a volume and makes the interface available on port 3000. The command below is the one from the Open WebUI documentation:
docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data -e WEBUI_SECRET_KEY=your-secret-key --name open-webui --restart always ghcr.io/open-webui/open-webui:main
| Method | Command | Default port |
|---|---|---|
| Docker | docker run … ghcr.io/open-webui/open-webui:main | 3000 (to 8080 inside the container) |
| Python (pip) | pip install open-webui then open-webui serve | 8080 |
Steps:
- Install Docker on the server.
- Run the command, replacing
your-secret-keywith a long random value. - Open the interface through an SSH tunnel or your reverse proxy.
- Create the admin account on first sign-in.
How do you connect Open WebUI to Ollama?
Install Ollama on the same machine, let it listen locally on port 11434, then enter its address in the Open WebUI connection settings. The --add-host=host.docker.internal:host-gateway option in the Docker command lets the container reach a service on the host machine.
- Install Ollama:
curl -fsSL https://ollama.com/install.sh | sh, the command from the official repository. - Pull a model:
ollama pull mistral. - In Open WebUI, open the admin settings, connections section.
- Enter the Ollama address:
http://host.docker.internal:11434. - Pick the model from the list and start a chat.
Pro tip: never expose Ollama to the Internet to connect it to Open WebUI. Both run on the same server: only the web interface, protected by HTTPS and accounts, should be reachable from outside.
Can you use Open WebUI with an OpenAI-compatible API?
Yes, Open WebUI accepts any OpenAI-compatible API in addition to Ollama: you just add the API address and a key in the connections. That lets you run the interface on a small server without a GPU and hand off the computing.
- Interface on a VPS: a VPS is enough to host Open WebUI on its own.
- Computing through an API: the Wayhost GPU page presents an OpenAI-compatible AI API, listed as usable with Open WebUI, billed per token from prepaid credit, with requests not stored (checked on November 6, 2026).
- A mix of both: a local model for sensitive documents, an API for everything else.
How do you manage users and documents?
Each person has an account, the administrator decides who can sign up and which models are visible, and documents are uploaded in the interface to query their content. That is what turns a personal tool into a team assistant.
- Sign-ups: turn off open registration and create accounts yourself.
- Roles: keep administrators and users separate.
- Documents: search across files (RAG) is one of the features described in the documentation.
- History: conversations are stored in the data volume, which should be backed up regularly.
What server do you need for Open WebUI?
The interface itself is light; the model dictates the machine: you need a GPU whose VRAM holds the chosen model, plus headroom for simultaneous requests. Model sizes are detailed in our Ollama guide.
| Use | Target model | Recommended machine |
|---|---|---|
| Trial, a few people | 7B to 8B parameters | Card with 8 GB of VRAM or more |
| Team, finer answers | 12B to 27B parameters | 32 GB GPU |
| Several models or a large one | 70B parameters | 2 GPUs or more |
| Interface only, remote computing | Remote API | VPS without a GPU |
- Simultaneous users lengthen response times: test with your actual team.
- Wayhost GPU rental offers a dedicated V100 32 GB at €0.60 excl. VAT per hour or €199.99 excl. VAT per month, in Cergy (checked on November 6, 2026).
- For a whole machine, see the physical server offer.
How do you secure Open WebUI in production?
Put Open WebUI behind an HTTPS reverse proxy, set a secret key, close open registration and keep Ollama listening locally only. An internal assistant quickly holds sensitive information.
- HTTPS: a reverse proxy (nginx, Caddy) with a certificate on your domain name.
- Secret key: the
WEBUI_SECRET_KEYvariable in the Docker command, long and random. - Accounts: closed registration, strong passwords, accounts removed when someone leaves.
- Firewall: only ports 80 and 443 open; 11434 and 3000 closed from outside.
- Backups: copy the
open-webuivolume and test restoring it. - Updates: pull the new Docker image regularly.
For the hosting itself, our guide to a cheap, sovereign VPS lists the sovereignty criteria to check.
How much does a self-hosted internal assistant cost?
The cost comes down to the server: Open WebUI and Ollama are free, and you pay for the GPU by the hour or by the month, however many users you have. It is the opposite of a per-person subscription.
- Software: Ollama uses the MIT license; Open WebUI is free under its own license.
- GPU server: a monthly plan for a permanent service, or hourly for trials.
- Admin time: updates, backups and accounts to plan for every month.
- Alternative: interface on a VPS and per-token API computing if usage is light.
Frequently asked questions
Is Open WebUI free?
Yes, Open WebUI can be downloaded and used for free, under its own license. The only cost is the server that hosts it. The models used through Ollama are free too, each with its own license.
Does Open WebUI need the Internet?
No, Open WebUI is designed to run entirely offline with local models. The Internet is only needed to pull the Docker image, models and updates. Conversations stay on your server.
Do you need a GPU for Open WebUI?
The interface itself does not need one. The model is what requires a GPU, through Ollama or another engine. Without a GPU, connect Open WebUI to a remote OpenAI-compatible API.
What is the difference between Open WebUI and Ollama?
Ollama runs the model and exposes an API. Open WebUI is the web interface that connects to it, with accounts and history. The two are used together on the same server.
Can you give a whole team access?
Yes, Open WebUI handles several accounts with roles. The administrator creates accounts and chooses which models are visible. Size the GPU for the number of simultaneous requests.
Can Open WebUI read my documents?
Yes, you can upload documents and ask questions about their content thanks to retrieval-augmented generation (RAG). The files stay on your server. Quality depends on the model and on how the documents are split.
How do you update Open WebUI?
With Docker, pull the new image, stop the old container and start it again with the same volume. Data and accounts are kept in the volume. Back up before every update.
Key takeaways
- Open WebUI is a self-hosted ChatGPT, offline, with accounts and history.
- It connects to Ollama locally, or to an OpenAI-compatible API.
- Docker installs it in one command, with the interface on port 3000.
- Choose the GPU for the model: 32 GB of VRAM for a 27B model, which weighs 17 GB in the Ollama library.
- HTTPS, a secret key, closed registration and a local-only Ollama are the basic production rules.
To run Open WebUI and Ollama on an NVIDIA V100 32 GB hosted in France, discover Wayhost GPU rental.
Article written by Mouhamed BANKOLE, CEO at ITNET Technologies.