MMOOZZGG.com

How to build an AI agent

How to install Open WebUI on your server and open it to staff

7 min read11 sources

Open WebUI is a chat with AI in the browser that a company puts on its own server and connects to its own models. To install Open WebUI you run one Docker command, then the administrator connects Ollama, uploads documents and opens access to staff. Since version 0.6.6 the program has its own license: with more than 50 users in 30 days, you may not remove the Open WebUI brand.

On this page · 6 sections

Build it yourself in a week

What is Open WebUI

Open WebUI gives staff a chat window in the browser, while the models run on your server. A separate program, usually Ollama, runs the models themselves, and Open WebUI adds accounts, chat history, knowledge bases and access rights on top of it.

For a company, this is a way to give people AI without sending their conversations to an outside service. Open models for such a chat exist: DeepSeek released R1 on 20 January 2025 under the MIT license, and the Qwen 3 family came out on 29 April 2025 under the Apache 2.0 license.

By itself Open WebUI is not an agent but a chat. It becomes an agent when the model has instructions, your documents and tools. Below we build such an assistant following the official documentation at docs.openwebui.com.

Installing with Docker

Docker is a program that runs an application in a ready container with all its dependencies, so you do not need to install Python and libraries by hand. Install Docker on the server or computer where the chat will run.

1. Generate a secret key with the command openssl rand -hex 32. You will need it in the next command in place of the words your-key.

2. Start Open WebUI with the command docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data -e WEBUI_SECRET_KEY=your-key --name open-webui --restart always ghcr.io/open-webui/open-webui:main. The command downloads the image and starts it by itself.

3. If the server has an Nvidia graphics card, take the image with the cuda tag and add the --gpus all flag to the command. If Ollama is not installed yet, there is an image with the ollama tag: the model and the interface in one container, running on the processor only.

4. In your browser, open http://localhost:3000. The first start takes about a minute. If the page does not open, wait for the line Application startup complete in the container log.

5. On the first screen, click “Create Admin Account”. The first account becomes the administrator: it manages users and all settings.

Without Docker, you install Open WebUI with the command pip install open-webui and start it with open-webui serve, after which it opens on port 8080. Python 3.11 or 3.12 works; version 3.13 is not supported yet.

A model through Ollama

Ollama runs open models on your computer or server and is released under the MIT license. It usually listens on port 11434, and if it is on the same server as the Open WebUI container, its address is http://host.docker.internal:11434.

1. Open “Settings”, then “Admin” and “Connections”. Find your Ollama server in the “Manage Ollama API Connections” list.

2. Click “Manage” (the download icon) next to the connection and choose “Pull a model from Ollama.com”. The model downloads right from the interface, without a terminal.

3. Choose the model in the model list above the chat. If you type the name of a model you do not have yet, Open WebUI offers to download it.

How much memory a model needs and which server to take for it, we covered on the page about servers for AI. We compared programs for running models separately, and the beginner’s path without a server is described on the page about a local agent.

Knowledge base and your own assistant

A knowledge base is a set of your documents in which the model looks for the right pieces before it answers. This is how RAG works: the model answers from the text it found, not from memory, and you do not need to train it for this.

1. Open “Workspace”, then “Knowledge”, and click “Create”. Give the base a name and a description.

2. Click “+” next to the search, choose “Upload files” in the “Add Content” menu and upload the documents. The documentation suggests starting with a small text PDF and waiting until processing finishes.

3. Test the base in an ordinary chat: type the # symbol and choose the base, then ask a question about the document.

4. Build an assistant: “Workspace”, “Models”, “Create”. Enter a name, choose a model in the “Base Model” field, write the rules in “System Prompt”, connect the base in “Knowledge” and click “Save & Create”.

Such an assistant already has instructions and knowledge. Tools, that is, actions in external systems, are chosen in the “Tools” field of the same editor, and in “Capabilities” you turn on, for example, web search and code execution.

Access for staff

After the first account is created, sign-up for new users is turned off. New accounts wait in the “Pending” status and see no models until the administrator approves them.

1. Open “Settings”, “Admin”, “Authentication”. To let staff sign up by themselves, turn on “New Sign Ups”; the value user in “Default User Role” lets them in without waiting for approval.

2. Create groups by department and give them access to the right models and knowledge bases. Permissions in Open WebUI add up, so check the general settings and group permissions together.

3. For a knowledge base, access is changed in the “⋮” menu of its row, the “Access” item. For an assistant, it is the “Access” button in the editor: access for individual people, groups or everyone.

4. Log in as an ordinary user from the right group and check that closed models and bases are not visible to them.

For work from other computers, the documentation recommends HTTPS: Open WebUI is placed behind a reverse proxy, such as Nginx, on the company’s domain.

Staff will upload contracts and customer conversations to the chat, and that is personal data. For a Russian company, 152-FZ, Russia’s personal data law, requires collecting and storing it in databases in Russia, and since 1 March 2023 the operator must notify Roskomnadzor, the data protection regulator, before a transfer abroad. Keep the Open WebUI server and the models in Russia; more on the page about data security.

License: the brand condition

Open WebUI code up to and including version 0.6.5 is released under the BSD-3-Clause license. From version 0.6.6 of 19 April 2025, the Open WebUI License applies, and the documentation itself says that it is not an open license in the OSI sense.

You may use it widely: install it on your own servers, modify the code and the interface, give it to your team and even charge for access. There is one condition: you may not change, remove or hide the Open WebUI brand, that is, the name, logo and marks in the interface.

You may change the brand in three cases: you have no more than 50 users in 30 days, you have an enterprise license from the developers, or you are a notable contributor to the project with their written permission. For a company’s internal chat the rule is simple: the Open WebUI logo stays in place.

Questions and answers

Can a company use Open WebUI for free?

Yes, the license allows you to install Open WebUI on your own servers, modify it and give it to your team. The condition concerns the brand: if you have more than 50 users in 30 days, you may not remove the Open WebUI name and logo without an enterprise license. You pay for the server and the administrator’s time.

How is Open WebUI different from Ollama?

Ollama runs the model and answers requests from programs. Open WebUI is the interface for people on top of it: accounts, chat history, knowledge bases and access rights. Together they give you an internal AI chat on your own server.

Sources and fact-checking

Checked by the editors on 16 October 2026.

  1. Open WebUI Docs: License, 19 April 2025 (docs.openwebui.com)
  2. Open WebUI Docs: Quick Start, 2026 (docs.openwebui.com)
  3. Open WebUI Docs: Starting with Ollama, 2026 (docs.openwebui.com)
  4. Open WebUI Docs: Share Open WebUI with Your Team, 2026 (docs.openwebui.com)
  5. Alpine Linux packages, 2026 (pkgs.alpinelinux.org)
  6. Simon Willison, 20 January 2025 (simonwillison.net)
  7. Simon Willison, 29 April 2025 (simonwillison.net)
  8. arXiv 2005.11401, 22 May 2020 (arxiv.org)
  9. Brace Law Firm, 9 July 2023 (brace-lf.com)
  10. Open WebUI documentation: Knowledge, checked 16 October 2026 (docs.openwebui.com)
  11. Open WebUI documentation: Models, checked 16 October 2026 (docs.openwebui.com)

Found a mistake? Tell us

Publications

Contact us