Download websites that use Sphinx

About Sphinx

Start with a warning that matters more than any statistic on this page. The name Sphinx belongs to at least two well known pieces of software: a full text search server used behind database driven sites, and a documentation generator from the Python world that stamps its identity into every page it builds. Our record holds a single Python value for the name, and no single value covers both projects. The two leave traces on the public web, and traces are what we index.

We publish what we detect and label it under Documentation rather than pretend the ambiguity away. If your question concerns search infrastructure, filter the export against the other signals recorded on the same domains, because a documentation site and a search backend look nothing alike once you examine the rest of the stack. Stating the caveat costs us a cleaner sales line and saves you a wasted afternoon, which is the correct trade for data somebody is going to act on.

With that said, the scale and the version data are real. As of September 2026 we see Sphinx on 2,126 sites. We have recorded 123 distinct versions, 9.1.0 among them, which is a wide spread for something this specialised and tells you that installations are pinned rather than continuously updated. Software that ships as a package, gets configured once and then sits still produces exactly this pattern.

Our record dates the entry to 2008 under a BSD-2-Clause licence, and both values describe the search server rather than the documentation tool whose site we link to, which is the same ambiguity turning up in the fields themselves. The search side belongs to an older generation of infrastructure, the era when adding real search to a site meant running a separate daemon next to MySQL or PostgreSQL rather than calling a hosted API. Some of that generation has quietly disappeared. Some of it is load bearing inside systems nobody wants to touch, and the second group is the interesting one, because it is where migration work, support contracts and long tail risk actually live.

Detection of server side software is always partial. A search daemon exposes itself only when the application in front of it says so, through error strings, response headers, generated markup or a footer nobody edited, and a carefully built application says nothing at all. The count is therefore a lower bound drawn from the sites that leak. What the downloadable list gives you that a single lookup cannot is the shape of that population: which domains, which versions, which top level domains, and what moved between one export and the next.