SearX

Last reviewed Aug 26, 2026 · 750 words

SearX is finished. The original repository was archived in 2023 after its maintainer wound the project down, and no security fixes or engine updates have shipped since. If you are about to install it, install SearXNG instead: same idea, same AGPL-3.0 licence, actively maintained, and a superset of the configuration. If you already run SearX, the rest of this page is about why it is quietly breaking and how to migrate without losing your settings. The 13,538 stars on the archived repository are a monument, not a signal that it is safe to use.

What SearX got right in 2014

Adam Tauber's SearX was the first widely used metasearch engine built around not logging the user. It sends your query to a configurable list of upstream engines (Google, Bing, DuckDuckGo, Wikipedia, and dozens of niche sources), merges and re-ranks the results, and returns them without cookies, tracking parameters, or a record of who asked. Because the upstream engines see the instance's IP rather than yours, a private instance is a real privacy gain: Google sees one anonymous server making queries, not your browser fingerprint. The JSON API (?format=json) made it scriptable, and the engine list made it extensible. Every privacy search tool since, including Whoogle and 4get, works in its shadow.

Why an unmaintained metasearch engine rots faster than most software

A metasearch engine is a collection of scrapers, and scrapers break when the site they scrape changes its HTML. Google changes result markup several times a year; Bing and the others less often but still regularly. SearXNG ships engine fixes within days because it has an active maintainer group. SearX has shipped none since it was archived, so an old instance today returns empty or partial results from its most important engines, and the failure looks like "search is slow and finds nothing" rather than an error. Add the unpatched Python dependencies underneath and the picture is clear: it is not that SearX is bad, it is that this category of software cannot survive without constant upstream maintenance.

Migrating to SearXNG keeps most of your settings

SearXNG forked from SearX in 2021 and kept the settings.yml format, so migration is mostly a copy with three edits.

# settings.yml (SearXNG)
use_default_settings: true
server:
  secret_key: "generate-a-new-random-one"
  limiter: false
search:
  formats:
    - html
    - json

Generate a fresh secret_key (openssl rand -hex 32), set use_default_settings: true so you only override what you change, and re-enable the JSON format explicitly if anything scripts against it, because SearXNG disables it by default. The Docker image is searxng/searxng and listens on 8080 rather than SearX's 8888, so update your reverse proxy. Engine names are largely the same; a few SearX engines were dropped and many were added, so review the engines: list in the SearXNG defaults rather than pasting your old one wholesale. The whole move is under an hour for a single-user instance, and the 256 MB the catalogue lists for SearX is about what SearXNG idles at too.

Private instance or public instance

A private instance for a household works well: a few hundred queries a day from one IP looks like a normal user to the upstream engines. A public instance does not: once query volume from one IP climbs, Google and Bing start serving CAPTCHAs or blocking, and results degrade for everyone. SearXNG's limiter (which needs a Valkey or Redis instance) exists to slow abusive clients on public instances; leave it off for private use. If you want a public service, expect to rotate outbound IPs or accept degraded Google results, and read the project's documentation on running a public instance first. The broader search category has the alternatives, including tools that build their own index rather than borrowing someone else's.

What I'd do

Run SearXNG behind Caddy on a small VPS or a home box, JSON format enabled, limiter off, Google and DuckDuckGo as the primary engines with Wikipedia and a couple of niche sources turned on, and set it as the browser default with bang shortcuts for the odd query that needs a specific engine. If you found this page because you still run SearX, migrate this week; every month the results get worse and the unpatched dependencies get older. For the "why bother at all" question, the Google Search alternatives page makes the case; for the how, this is it.