Open WebUI Web Search Not Working? It’s Off By Default (and how to fix it!)
Summary: Open WebUI ships web search off, and there are two separate ways to turn it on. For one chat: click the four-diamond Integrations icon right of the + and flip Web Search on. For every chat: Settings > Interface > Web Search in Chat, set it to Always, then Save. Neither one is in Admin Settings, which is where I went looking first.
Verified on Open WebUI 0.11.0, August 21 2026.
Option 1: Turn It On For One Chat
Here’s the whole message box. Look how little there is to go on. 👇🔍
- Click the four-diamond icon immediately right of the
+. Tooltip says Integrations. - You get two switches, Web Search and Code Interpreter, both showing off.
- Flip Web Search on.
- Send your prompt like normal.
Your confirmation is that blue globe that shows up in the icon row next to the diamond. The diamond itself stays gray, so don’t sit there waiting on it to change color like I did 😅
The downside is right there in the name. It’s per chat, so unless you also do Option 2 below, every new chat starts back at off.
And if you found a guide saying the toggle is under the + button, that guide is written for an older release. 0.11.0 rebuilt the interface and moved it. (and be forewarned, future versions may change it again!)
Option 2: Turn It On For Every Chat
This is the one I’d go with, and it’s buried quite deep lol.
Click your avatar in the bottom left corner, then Settings. Not the Admin Panel, Settings. Then Interface in the left column.
Now scroll way down. Past Stylized PDF Export, past Floating Quick Actions, almost to the Input section. Web Search in Chat is likely sitting down there at Default. Click it over to Always and hit Save.
Its own description says it plainly enough: “Set web search availability for new chats.”
Now here’s the part that’ll have you convinced it didn’t work.
With Always set, the Web Search switch in the Integrations menu still shows off. Search is on the table anyway. I tested it with that switch visibly off and Qwen3.8 went and pulled 5 sources on a weather question. So trust the setting, not the switch.
Why It Looks Potentially Broken (instead of just not toggled ON)
If you have the toggles off you don’t get any true errors, you just may find in the thinking that there’s no web search available.
Here’s Qwen3.8-27B thinking out loud:
“there’s no web search tool here. The available tools are knowledge base search, chat, notes, calendar.”
Turns out the builtin tools it did list are gated on their own settings and need nothing sent per request. Web search also needs a per-request flag, and in the UI that flag comes from the chat toggle or the Always setting. Open WebUI’s own troubleshooting page names the three gates (global, model, chat) but isn’t very explicit about where to find the chat one 😢
To be fair to the model though, this isn’t a tool-calling failure. When search is toggled Qwen3.8 picks it up, 9 out of 9 in my testing. Which tracks with what I found when I tested 13 local LLMs on tool calling.
Why This Is Important
Once it was on, Qwen3.8 fired off six searches on its own and came back with 22 sources. 👍
One important thing to remember about current LLMs and their knowledge bases, is that they are frozen at a single point in time when trained. Whatever it learned, it learned once, then training stopped. Ask about something that shipped last month and it’ll either tell you that doesn’t exist or invent something that sounds right. It has no way of knowing what it’s missing.
Out of the box that’s the biggest practical gap between a current cutting edge 17 GB model (Qwen3.8-27B at Q4_K_M) running on hardware you own and the expensive frontier ones, and it isn’t really as much about intelligence as harnessing and tool calling (especially web, when it comes to research purposes). ChatGPT, Claude and Gemini all ship with web search built in. That’s a good chunk of why they feel current and your local model feels stale.
Wire up a backend once and web search closes most of that gap. It doesn’t quite make Qwen3.8-27B the equal of a frontier model, but it can get you quite close for many tasks AND you can even get BETTER web capabilities than some of the web based services that are blocked and or only use snippets (caveat, with local the models the sky is the limit with how you want your models to search and use the web, lots of options are out there and they range from meh to awesome).
You Still Need a Search Backend
Both options above only do something if you’ve got a provider configured under Admin Settings > Web Search. Mine’s a self-hosted SearXNG instance in its own container sitting right next to Open WebUI on a Minisforum MS-01 and that container calls another machine that actually handles running the LLMs on NVIDIA or AMD GPUs, depending on the model. For search Google, Bing, Brave and DuckDuckGo can all work too if you’d rather not self-host, though several of them want an API key. Setting up SearXNG is its own post.
NOTE: Using LM Studio for chat instead? LM Studio has no built-in web search at all, you install a plugin from the LM Studio Hub first. After that it acts like Open WebUI does and stays off until you turn it on per chat from the tools button in the chat input. So if your LM Studio model can’t search, confirm you ever installed the plugin before you go digging anywhere else.
Conclusion:
Two options for enabling web search, in two completely different places, and out of the box neither one is enabled. For most use cases my recommendation is to set Web Search in Chat to Always. Enjoy your LLM and happy web searching! 👍👍
Sources and Further Reading
- Open WebUI v0.11.0 release notes
- Open WebUI blog: v0.11.0, The Interface Reorganized
- Open WebUI docs: Web Search troubleshooting
- LM Studio Hub: web-search plugin
Information in this post was accurate when published on August 21, 2026 and was verified on Open WebUI 0.11.0. Software interfaces change.