Open WebUI for Teams: User Approval, Groups, Custom AI Assistants, Prompt Library, Shared Knowledge and Backups

After installing Open WebUI with Ollama, most teams use it like a private ChatGPT: everyone opens a chat and types. It becomes far more useful when you set it up for the team: approve accounts, put staff into groups, build ready-made assistants for repeated tasks such as customer replies, translation and product copy, share a prompt library and company documents, and back everything up. This guide walks through those settings. Menu names follow recent Open WebUI versions and occasionally move between releases.

Step 1: Control Who Can Sign Up

The first account created becomes the administrator. Then, under Admin Panel > Settings > General:

• Set the default role for new users to pending, so new sign-ups cannot chat until an administrator approves them.
• Turn new sign-ups on only while colleagues are registering, and off again afterwards. You can also create accounts yourself under Admin Panel > Users.
• Keep the number of administrators small; everyone else gets the user role.

Step 2: Organise Staff into Groups

Under Admin Panel > Users > Groups, create groups such as Support, Marketing and Developers and add the members. Groups are used in two ways: to decide which assistants and knowledge each team can see, and to grant permissions such as uploading files or creating their own assistants and prompts in the workspace. A common setup is to let team leads edit the workspace and give everyone else chat access only.

Step 3: Decide Which Models Users See

Under Admin Panel > Settings > Models, hide models that should not appear in the chat menu, such as the embedding model or models you are only testing, and set a sensible default model. Staff then choose from a short list of purpose-built assistants instead of raw model names.

Step 4: Build Ready-Made Assistants

An assistant is a saved combination of a base model, a system prompt, settings and optional knowledge. Create them under Workspace > Models, give each one a clear name and description, and share it with the groups that need it. Some examples that work well:

Name: Customer Reply Writer        Base model: qwen2.5:7b
System prompt:
You help support staff answer customer messages for ACME Hosting.
Write a polite, concise reply in the customer's language.
Use only facts given by the staff member or the attached knowledge.
If information is missing, list what to ask the customer instead of guessing.

Name: Listing Translator           Base model: qwen2.5:7b
System prompt:
Translate e-commerce product text into the language the user requests.
Keep brand names, model numbers, units and HTML tags unchanged.
Use natural marketplace wording, not word-for-word translation.
Return only the translated text.

Name: Ticket Summariser            Base model: qwen2.5:7b
System prompt:
Summarise the conversation pasted by the user in 5 bullet points:
problem, what was tried, current status, next step, customer mood.

For assistants that must stick to facts, lower the temperature in the advanced parameters (for example to 0.3). Test each assistant with real examples before you share it.

Step 5: Share Company Knowledge

Under Workspace > Knowledge, create collections such as "Product FAQ" or "Support procedures" and upload documents. Attach a collection to an assistant so every chat with it can use those documents, or type # in a chat to pick a collection or a web page for a single question. Keep collections current: delete outdated files when prices or procedures change. For a stricter document question-and-answer setup, see building a private RAG chatbot with AnythingLLM.

Step 6: Create a Prompt Library

Under Workspace > Prompts, save instructions your team types again and again, each with a short command. In any chat, typing / lists them. For example:

Command: /seo-title
Content:
Write 5 product titles for the product below. Each title must be under
120 characters, put the main keyword first and make no exaggerated claims.

Command: /polite
Content:
Rewrite the text below so it sounds friendly and professional.
Keep the meaning and all facts unchanged.

Step 7: Daily Tips for Staff

• Type @ to hand the next answer to a different assistant without leaving the conversation.
• Type # to use a knowledge collection or a URL, and / to insert a saved prompt.
• Upload a file into a chat to ask questions about that document only.
• Select two models at the top of a chat to compare their answers side by side.
• Use a temporary chat for anything that should not be saved in the history.

Step 8: Back Up and Update

All users, chats, assistants, prompts and uploaded files live in the open-webui Docker volume. Back it up before every update:

# Back up the whole Open WebUI data volume
mkdir -p /root/backup
docker run --rm -v open-webui:/data -v /root/backup:/backup alpine \
  tar czf /backup/open-webui-$(date +%F).tar.gz -C /data .

# Update to the latest version (the data stays in the volume)
docker pull ghcr.io/open-webui/open-webui:main
docker stop open-webui && docker rm open-webui
# then start it again with the same "docker run" command you used to install it

FAQ

How many people can use it at the same time?

Open WebUI itself is light; the limit is how fast the model generates text. On a CPU server, answers are produced roughly one after another, so busy periods mean waiting. Use smaller models for simple tasks, keep the most-used model loaded, or add a cloud model for heavy use as described in the next question.

Can I add cloud models next to the local ones?

Yes. Under Admin Panel > Settings > Connections you can add any OpenAI-compatible API with its base URL and key. Use providers only from regions they support and within their terms, and remember that prompts sent to those models leave your server.

A colleague forgot their password

An administrator can set a new password by editing the user under Admin Panel > Users.

Still stuck after following these steps? Open a support ticket and the IMIDC 24/7 technical team will help. Include the server IP, the Open WebUI version, the output of docker logs open-webui and the full error message so we can pinpoint the issue faster.