SearXNG Puts a New Interface on Google and Bing Results Without Replacing Them
The first thing to know is that SearXNG does not crawl the web. SearXNG is a metasearch engine, not a search engine. It owns no index. Wikipedia's entry puts it simply: as a metasearch engine, SearXNG functions by sending queries to upstream search engines and returning them to the user. Your results still come from Google, Bing and the other services, and their quality, freshness and blocking rules still apply.
The XDA writer's main complaint is Google's cluttered results page. That now means an AI overview at the top, plus image, definition and question-and-answer widgets. SearXNG returns a standard list of links. Some links have thumbnails, and an info panel sometimes appears. Images and other kinds of results sit on separate tabs. Wikipedia lists the categories as "General," "Images," "Videos," and "News," as well as the non-standard categories of "Social Media," "Music," "Files," "IT," and "Science." The XDA author thought IT and Science were odd choices for dedicated tabs.
SearXNG also labels each result with the engine or engines that returned it. That is useful in practice: after a few weeks you can see which engines bring back useful results for the kinds of things you search for, and which you could turn off. Whether the interface is actually better than Google's is a matter of taste. The labelling is a real feature either way.
How SearXNG Strips Cookies and Browser Fingerprints From Upstream Requests
SearXNG's documentation describes three protections. It removes private data from requests sent to search services. It does not pass along ads or other third-party content from those services. And it removes private data from requests to the result pages you click through to. In practice, SearXNG sends no cookies to the external engines and creates a random browser profile for every request. Google sees the SearXNG server's IP address, not yours, and the software can be set up to route traffic through a proxy or Tor.
The documentation adds that SearXNG serves no ads or tracking content, and that websites you open from a results page can't see which page sent you or what you searched for. Privacy consultant Michael Bazzell makes the same point in his IntelTechniques guide: with a public instance, all queries are submitted to search engines from a third-party server. The IP addresses collected from engines are those of the server, not yours.
There is a limit, and it matters most for self-hosters. Google and Bing still receive your search text, just without your cookies. If you run a private instance from home, the IP address those engines log is your home connection's. Bazzell describes self-hosting this way: all queries are submitted from your machine directly to the engines. The tracking code on engine websites is removed from the SearXNG pages. You avoid cookies, tracking scripts and the engines' own account profiles. You don't become invisible on the network unless you add the proxy or Tor option.
Google has also pushed back against metasearch traffic before. According to Wikipedia, in 2019, Google has begun to block some self-hosted instances. In response, some instances have been modified to silently skip trying to search with Google even when it's the only engine specified. SearXNG itself forked from searx in the middle of 2021, with a view to provide faster debugging and fixes of engine errors.
Engine Toggles, settings.yml and Bang Shortcuts Give SearXNG Users Control
Choosing your engines is the feature the XDA piece praises most, and SearXNG gives you three ways to do it.
The first is the Preferences page, where you can switch individual engines on or off. XDA says its default setup included Google, Bing, DuckDuckGo and Qwant for general searches, Wikipedia, and translation engines such as dictzone and lingva. If you don't want Google to see your queries at all, you can turn it off there.
The second is the ! prefix, which sends one search to specific engines or categories. SearXNG's search syntax documentation uses !wp paris to search Wikipedia and !images Wau Holland for an image search. Modifiers can be combined and add to each other, so !map !ddg !wp paris searches the map category plus DuckDuckGo and Wikipedia. A : prefix sets the language, as in :fr !wp Wau Holland. XDA's example is !wq for Wikiquote. It says shortcuts work even for engines you have turned off in your general searches, so you can keep your default results tidy and still reach niche sources when you need them. Shortcut names depend on how an instance is set up, so check that instance's Preferences page for the list.
Don't confuse the single ! with the double !!. SearXNG supports DuckDuckGo's external bangs, so !!wfr Wau Holland jumps straight to an outside search page. The documentation says that in this case the search runs directly on the external engine, and SearXNG cannot protect your privacy. The same goes for a standalone !! in a query, which takes you straight to the first result, much like "I'm Feeling Lucky." If privacy is why you use SearXNG, a double bang skips the protection you set it up for.
The third option, for server operators only, is the settings.yml file. SearXNG loads it from the path in the SEARXNG_SETTINGS_PATH environment variable or from /etc/searxng/settings.yml. If neither file exists, is empty or can't be read, it uses the built-in default. Setting use_default_settings: true lets you change only what you need. Under engines: you can add remove: to drop specific engines, such as Google, or keep_only: to run only a named list. You can also switch on an engine that is off by default, or add an access token for a private engine. The documentation's example adds a token for the Arch Linux wiki engine. XDA says it edited the file inside its container. Where the file lives, and whether edits survive an update, depends on how you installed SearXNG.
Engine counts vary by source. XDA says SearXNG includes 260 engines, not all of them switched on by default. A July 2026 self-hosting guide by Joshua Opolko says it forwards your query to up to 262 upstream engines, and a Bitdoze guide from about three weeks ago puts the figure at up to 272 search services. The number changes with each release, and any single instance offers only what its operator has enabled. Treat "about 260 or more" as a rough figure.
Public SearXNG Instances Trade Setup Work for Trust in a Stranger
You don't need to run a server to try SearXNG. A list of public instances is available at searx.space, and Bitdoze counted about 74 online right now. XDA notes that search preferences on a public instance are stored in your browser, so each person gets their own settings. You can't edit the server's settings.yml, though, so you are limited to the engines the operator has enabled.
SearXNG's own documentation is frank about the risk. If you use a public instance, you have to trust its administrator, and you can't know whether your searches are logged, aggregated, or passed or sold to someone else. Its advice is to use an instance run by people you trust. Bazzell thinks deliberate snooping is unlikely. He writes that this cannot be done with the stock SearXNG software, and hosts would have to go out of their way to collect data about users. Unlikely still means possible, and a stranger's server can always be changed.
Popular public servers also run into practical problems. SearXNG warns that badly protected public instances attract abuse, which can lead upstream engines to demand CAPTCHAs or ban the instance's IP address, and you get fewer results as a result. Bazzell lists the same risks: if the host is popular, some engines may block access. If the host has an outage, you are without service. Bitdoze adds rate limiting, JSON API disabled, someone else's engine selection, and no guarantee the operator is not logging queries.
XDA suggests it may be easier to trust a small server with no advertising than a company like Google. That is a fair argument, but it's still a judgment about trust, and each reader has to make it.
Self-Hosting SearXNG Moves the Maintenance Burden Onto You
Running your own instance removes the unknown operator. SearXNG's documentation says that on your own instance you control the source code, the logging settings and your private data. Because you can change the default settings on the server, your preferences don't have to live in cookies, and they won't reset when you clear your browser.
XDA installed SearXNG with a Proxmox helper script (Proxmox VE is a popular home-lab virtualization platform). It doesn't say which script or give the steps. Docker is the more common route. Opolko notes that the maintained 2026 path is the container/ compose files in the main repo; the old searxng-docker helper repo was archived on March 28, 2026. Default internal port is 8080. That matters if you are following an older tutorial that points at the separate searxng-docker repository. Opolko also advises you to add a real secret key and a reverse proxy before you expose it to the internet. The official configuration example shows secret_key next to a comment telling you to change it.
The resource cost is small. Bitdoze estimates that SearXNG + Valkey fits in 1-2 GB RAM. That is a $4-6/month VPS. If you plan to share the instance, Bitdoze says a private instance (limiter off, no public_instance flag) is fine for solo use. If you want to share it with others or make it public, you need the limiter enabled. Developers get something extra too. SearXNG has a JSON API that tools such as LiteLLM can call, though LiteLLM notes that SearXNG returns a fixed number of results per page (~20 by default) and does not support limiting results via the API.
What this means for you
Try a public instance first, and self-host only once you know you like the results. Picking an instance from searx.space costs nothing, and within a week you'll know whether SearXNG's plain list suits the way you search. If it does, and privacy is the reason, a private instance on a home server, a Proxmox box or a cheap VPS takes away the need to trust a stranger. You then have to keep it updated, secured and working when upstream engines block it. If you mainly want a less cluttered page, a trusted public instance may be all you need.
- SearXNG has no index of its own, so the quality of your results depends on which upstream engines are enabled and whether they are blocking or rate-limiting the instance.
- The single-
!shortcuts, such as!wpor!images, keep your search inside SearXNG, while double-!!external bangs send you straight to another engine with none of SearXNG's privacy protection. - On a public instance, the operator can in principle see your searches, so choose one run by people or organizations you have reason to trust.
- On your own server, you control which engines run by using the
remove:orkeep_only:lists insettings.yml, and you should set a realsecret_keybefore exposing the instance. - A home-hosted instance hides your cookies and browser fingerprint from Google, but the engines still see your home IP address unless you set up the proxy or Tor option.
- Tutorials that use the old searxng-docker repository may be out of date, since Opolko reports it was archived in March 2026 in favour of the container files in the main project.
XDA's switch holds up. What SearXNG really offers is choice over which engines see your searches and who runs the server in between, and a cleaner results page on top of that. Google's own results page keeps adding AI-generated content, so a front end you control may appeal to more people. In return you take on the job of running and trusting that front end, whether it's your own server or someone else's.