[Guide] Connecting Open WebUI+Ollama -> SearXNG

Setting up Open WebUI with web search

Force your clanker to search the web through your own SearXNG.

The settings you need are spread across three different screens. That’s the part that catches people out, so do this in order.

1. Install and start the services

Install Ollama, SearXNG and Open WebUI, and start all three.

Open SearXNG’s Web UI and run a search, just to confirm it works on its own before anything depends on it.

2. Create your admin account

Open Open WebUI’s Web UI and register your admin account.

Do this before anything else. The Configure Backends action won’t run until an admin account exists.

3. Connect Ollama

In StartOS, open the Open WebUI service, go to Actions → Configure Backends, check Ollama, and save. Open WebUI restarts.

Then reload the Open WebUI page in your browser. It won’t pick up the new connection otherwise.

4. Download a model

Open a chat, click the model dropdown, and type a model name — qwen2.5:3b if you can afford the resources, llama3.1:1b if you want something really simple or are just testing this guide. When nothing matches, it offers to pull that model from Ollama.com. Click it and wait for the download.

Open WebUI doesn’t do the downloading itself; it asks Ollama to, and the model lands in Ollama’s storage.

If your server has no GPU, stay small.

5. Three settings, three places

Admin Settings → Web Search – turn web search on. The engine (searxng) and the query URL are already filled in for you. Leave the URL alone because it’s managed and kept correct automatically.

Settings → Interface → Web Search in Chat — set this to Always. Without it, web search is available but switched off in every new chat, so nothing searches and nothing tells you that.

Settings → General → Advanced → Function Calling — set this to Legacy if you’re using a small model.

That last one matters far more than it looks. On Native, Open WebUI offers the model a search tool and leaves the decision to it — and smaller models are too stupid to use it and there’s no indication as to why.

6. Confirm it works

Ask something that needs current information — “what’s the weather in Madrid today?” You should get an answer with sources listed underneath it. Those links are your proof.

To watch it from SearXNG, run SearXNG’s Config action and turn on the stats dashboard. The /stats page then counts every query it serves, so you can ask a question and watch the number change.

Different unrelated question…
is it possible to connect the dgx spark to the start9 and use the power from the dgx spark and interface with it via startos?
Thanks

Agreed, completely unrelated question and a strange place to ask it. Try here → https://community.start9.com/