Goal: use an already installed Open WebUI as the chat interface for an already installed Ollama model. This recipe configures the connection and checks a real response; it does not install software or change network exposure.
Before you begin
You need a running Ollama installation with at least one downloaded model, a working Open WebUI installation, and permission to manage its connections. Know whether each application runs directly on your computer or inside a container. A page opening in your browser proves only that the interface is reachable.
1. Confirm the model server
On the computer running Ollama, run ollama list and note an exact installed model name. In Windows PowerShell, run Invoke-RestMethod http://localhost:11434/api/tags. On macOS or Linux, use curl http://localhost:11434/api/tags. A response with a models array confirms this endpoint is reachable from that computer. If it fails, fix the Ollama connection first rather than changing the WebUI repeatedly.
2. Set the connection in Open WebUI
- Open the admin settings and select Connections. Find the Ollama connection settings, rather than an OpenAI-compatible provider entry.
- If both apps run directly on the same computer, use
http://localhost:11434. If Open WebUI runs in Docker and Ollama runs on its host, the documented address ishttp://host.docker.internal:11434, provided the existing deployment supports that host mapping and access. If both are on a container network, use the actual Ollama service name on that network. A bundled installation should retain its documented bundled configuration. - Use the connection verification control, then save the settings. Refresh the available models and select the installed model you noted earlier.
If the host mapping or listener is not already configured, stop here and follow the troubleshooting guide for your layout. Do not turn a private local API into a public endpoint, disable authentication, or alter firewall rules just to make this test pass.
3. Check a real conversation
Start a new chat with the selected local model. Ask: Give me a three-item packing list for a rainy walk. Then ask: Which item helps keep my feet dry? Check that you get an answer and that the follow-up uses the conversation. This is an inference test, not an accuracy benchmark.
4. Check what stays local
Keep cloud providers, external tools, web search and integrations disabled for this test. A local Ollama connection does not make every feature in the interface local. Chat history may be stored by the Open WebUI server, which is not necessarily the computer displaying the browser. After required downloads, an optional offline retest can confirm that this particular workflow operates without internet access; it does not audit all plugins or future changes.
Finished when
- The connection verifies and lists the intended installed model.
- A new chat produces a response and a coherent follow-up.
- You know where the model runs and where the chat history is stored.
Documentation checked November 14, 2026. This is a source-based connection recipe, not a claim of a fresh hands-on installation test.