Open WebUI + SearXNG: "404 Client Error: Not Found for url: https://example.com/search?...&number_of_results=3"

By MV2 of Munim, Inc., with AI agents. Checked on 11 November 2026 (about 00:03 UTC) by running the tool's own search code against example.com, a real local SearXNG and two unreachable addresses. Open WebUI v0.13.3 was the newest release then.

You set up SearXNG for web search in Open WebUI, ask a question, and the chat shows the search failing against example.com, as if your URL had been ignored. The report in open-webui#33995 (Open WebUI v0.13.3, Docker) showed:

{
  "error": "404 Client Error: Not Found for url: https://example.com/search?q=noticias+tecnolog%C3%ADa+hoy+2025&format=json&number_of_results=3"
}

When we ran the same code, the status line in the chat read:

Error during search: 404 Client Error: Not Found for url: https://example.com/search?q=open+webui&format=json&number_of_results=3

Short answer: Open WebUI does not fall back to example.com. This request comes from a community tool, almost always "Web Search using SearXNG and Scrape first N Pages" (by constLiakos, with enhancements by justinh-rahb and ther3zz). The tool has its own URL setting, SEARXNG_ENGINE_API_BASE_URL, and its default is the placeholder https://example.com/search. Changing the SearXNG URL in Admin Panel > Settings > Web Search doesn't touch it. Fix: open Workspace > Tools, click the gear icon (Valves) next to that tool, and set SEARXNG_ENGINE_API_BASE_URL to your SearXNG address including /search, for example http://searxng:8080/search. Or switch the tool off and use the built-in web search instead.

Tool or built-in search? How to tell from the message

Open WebUI's maintainer Classic298 pointed this out in #33995: the built-in search never sends number_of_results, and its errors look different. We checked both against their source code:

The community tool (0.1.12 and 0.1.14)Open WebUI's built-in SearXNG search (v0.13.3)
Where its URL comes fromOnly the tool's own valve: search_engine_url = self.valves.SEARXNG_ENGINE_API_BASE_URLAdmin Panel > Settings > Web Search (Query URL), or SEARXNG_QUERY_URL
Default URLhttps://example.com/searchnone
Parameters it sendsq, format, number_of_resultsq, format, pageno, safesearch, language, time_range, categories, theme, image_proxy
HTTP libraryrequests: messages read 404 Client Error: Not Found for url: ...aiohttp: per Classic298, messages read 404, message='Not Found', url='...'
Status lines in the chatInitiating web search for: ..., Sending request to search engine, then Error during search: ...Open WebUI's own web search status

So if the URL in the error ends in number_of_results=3 or the message starts with Error during search:, it is the tool. The model called it because the tool is enabled for that model or chat.

Every mistake and its exact message

We ran the tool's own search_web() function in a test, with a stand-in for Open WebUI that records the status lines, against example.com, an unused local port, a hostname that doesn't exist, and a real SearXNG (built from its official source, default engines, rate limiter off). Both versions of the tool we found printed exactly the same thing in every case. The query was open webui.

SituationValve valueWhat the chat shows (the tool's last status line)Fix
Valve never set(default)Error during search: 404 Client Error: Not Found for url: https://example.com/search?q=open+webui&format=json&number_of_results=3Set the valve to your SearXNG URL.
SearXNG reached, but JSON output not enabledhttp://127.0.0.1:8888/searchError during search: 403 Client Error: FORBIDDEN for url: http://127.0.0.1:8888/search?q=open+webui&format=json&number_of_results=3Add json to search: formats: in SearXNG's settings.yml and restart SearXNG.
Nothing listening at that address (wrong port, SearXNG down, or localhost inside a container)http://127.0.0.1:9/searchError during search: HTTPConnectionPool(host='127.0.0.1', port=9): Max retries exceeded with url: /search?q=open+webui&format=json&number_of_results=3 (Caused by NewConnectionError("HTTPConnection(host='127.0.0.1', port=9): Failed to establish a new connection: [Errno 111] Connection refused"))Use the address Open WebUI's server can reach (see below).
Hostname doesn't resolve (typo, or the containers aren't on the same Docker network)http://searxng.invalid:8080/searchError during search: HTTPConnectionPool(host='searxng.invalid', port=8080): Max retries exceeded with url: /search?q=open+webui&format=json&number_of_results=3 (Caused by NameResolutionError("HTTPConnection(host='searxng.invalid', port=8080): Failed to resolve 'searxng.invalid' ([Errno -2] Name or service not known)"))Use the SearXNG service or container name, and put both containers on one network.
Working, URL without /searchhttp://127.0.0.1:8888Web search completed. Retrieved content from 3 pagesWorked in our test, because SearXNG answered /?q=... with 308 PERMANENT REDIRECT to /search?q=... and the tool followed it. Adding /search is still the safer setting.
Workinghttp://127.0.0.1:8888/searchRetrieved 5 search results, Processing search results, Web search completed. Retrieved content from 3 pagesNothing to fix.

The JSON-off answer from SearXNG itself was HTTP/1.0 403 FORBIDDEN with an HTML "Forbidden" page. SearXNG ignored number_of_results=3 in our run: the tool still got 5 results back (its own SCRAPPED_PAGES_NO limit) and then read the first 3 pages.

Fix it

  1. Find the tool. Go to Workspace > Tools and look for "Web Search using SearXNG and Scrape first N Pages", or any tool with a SEARXNG_ENGINE_API_BASE_URL valve. At least one other published SearXNG tool, EntropyYue/web_search (version 10.4), uses the same https://example.com/search default. If you also have functions installed, check Admin Panel > Functions too.
  2. Set the URL. Click the gear icon (Valves) next to the tool and set SEARXNG_ENGINE_API_BASE_URL. The request is made by Open WebUI's server, not your browser, so use an address the server can reach: Inside the Open WebUI container, localhost and 127.0.0.1 mean that container, not your machine.
  3. Turn on JSON in SearXNG if you get the 403. In SearXNG's settings.yml (this is what our working test used):
    search:
      formats: [html, json]
    then restart SearXNG.
  4. Check from the Open WebUI container before trying the chat again: docker exec open-webui curl -s -o /dev/null -w '%{http_code}\n' 'http://searxng:8080/search?q=test&format=json' should print 200 (use your container name and URL; if the image has no curl, run python3 -c with urllib.request instead).

Or skip the tool: switch it off in Workspace > Tools (and in the model's tool list) and use the built-in search: Admin Panel > Settings > Web Search, engine searxng, Query URL with the search path, for example http://searxng:8080/search, as in Open WebUI's SearXNG guide. Then turn on web search in the chat. The built-in search needs the same JSON setting in SearXNG.

If you keep the tool, set the valve even if you never meant to use it: as long as the tool is on for a model, the model can call it, and it will send your users' search queries to example.com (which answers 404).

How this was made, and its limits

AI agents working for MV2 of Munim, Inc. saved two published copies of the tool on 11 November 2026: version 0.1.12 as posted in open-webui discussion #12990, and version 0.1.14 from its openwebui.com page (last updated 15 June 2026; the page showed 57,753 downloads and 137,859 views). Both copies have the same default URL and send the same parameters. They ran each copy's search_web() with Python 3.12 outside Open WebUI, with a small stand-in that records the status lines Open WebUI would show; that is the tool's own code, called in a test, not a full Open WebUI install. SearXNG was built from commit 57ff753 (9 November 2026), run once with only HTML output and once with HTML and JSON, with one query per case. The aiohttp message format for the built-in search is quoted from Classic298's reply in #33995; we read the built-in code at v0.13.3 but did not run it. The EntropyYue default was read from its source at commit 8569c87, not run. The tool is MIT-licensed and its authors' work; this page only explains its settings. Corrections are welcome as issues at github.com/munimtechnologies/local-ai-checkup.

Check the rest of your setup

The free local-ai-checkup script shows which AI ports on your machine other computers can reach, and the CVE checker tells you which security advisories affect your Open WebUI version. Both are read-only and send nothing off the machine.

Want someone to check it for you?

MV2 of Munim, Inc. does a $49 local AI health check for one machine, or a $199 small-office check for up to 5 machines or one shared Ollama or Open WebUI server. You get a written report of what's exposed, risky or slow and what to fix first, with exact commands for your OS. See a sample report. Pay $49 or pay $199 on Stripe, then email mv2@munimtech.com the output of python3 local_ai_checkup.py --json, your OS, and what you use it for. MV2 never asks for passwords, keys or remote access.