Is this the right next step?
Open WebUI is an interface, not the model doing the thinking. It sends requests to a model service you connect. Choose it when you want a browser-based workspace and are comfortable maintaining Docker. For the simplest first local chat, try LM Studio instead. This guide is documentation-checked as of November 14, 2026, not an installation or security audit.
1. Prepare the computer
The official quickstart supports Windows, macOS and Linux and recommends Docker. Install Docker using its official instructions for your operating system: Docker Desktop on Windows or Mac, or the appropriate Docker installation on Linux. Make sure Docker is running before proceeding. Model hardware requirements are separate from the interface: a cloud model reduces local compute requirements, while a local model still needs suitable memory and processing power.
Official installation options. For this introductory setup, use one computer, one user and no public domain. Sharing a service safely is a separate project, not simply opening a router port.
2. Create a persistent, local-only instance
The command below adapts the official Docker example by binding the interface to 127.0.0.1, so it is intended for this computer only. Replace REPLACE_WITH_RANDOM_SECRET with a long random secret generated by your password manager, and retain it privately. Do not publish the command containing your real secret.
docker run -d -p 127.0.0.1:3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data -e WEBUI_SECRET_KEY=REPLACE_WITH_RANDOM_SECRET --name open-webui --restart always ghcr.io/open-webui/open-webui:main
The named open-webui volume retains users, chats and settings separately from the container. Keep it. The main image follows current builds; a fixed release tag is preferable when you need a reproducible version. Open http://localhost:3000 and complete the initial account setup. Command options explained.
3. Connect one model and test it
- Open Settings → Admin → Connections.
- Add your working local model service, or a cloud provider using its documented URL and API key. Start with one connection so failures are easier to locate.
- Select a model in a new conversation. Ask it for a three-step packing checklist.
- Confirm you received an answer from the intended model. Then reload the page and check that your conversation remains.
A cloud connection sends the prompt and included context to that provider. Local inference does not make separately configured embedding, extraction or tool services local. Review model access and billing before assuming a chat subscription covers API calls. Provider connection and privacy guidance.
Connection problems
For Ollama, first verify it works outside Open WebUI. Inside a container, localhost refers to that container, not automatically to your host computer. The official Ollama connection guide explains host addressing and network troubleshooting. Do not expose Ollama publicly merely to fix a local connection. Connect Ollama.
Update without losing your workspace
Before upgrading, back up the data volume, record the running image version and retain your secret and configuration. Review release notes, especially database migrations. Recreate the container using the same volume rather than deleting the volume. Avoid volume-pruning commands as routine maintenance. A previous image is not a complete rollback if the database has migrated; preserve a matching backup. After updating, test sign-in, chat history and your model connection. Official update and backup procedure.