Marginalia Search

Self-hostable internet search engine that indexes the small, old and non-commercial web instead of ranking the biggest sites.

Open source alternative to:

Google Search

Marginalia Search is an open source internet search engine built for the parts of the web that general search engines deprioritise: personal sites, forums, documentation, hobbyist pages and old-fashioned text-heavy content. Instead of optimising for engagement or commercial intent, it ranks results around how much original, human-written text a page contains, so small and independent sites can surface at all.

Beyond the public instance at search.marginalia.nu, the codebase can be run as a white-label search engine. Barebones mode installs the engine without the Marginalia-specific GUI so you can crawl your own set of sites or sideload data you already have — WARC files, ZIM archives, StackExchange dumps or a plain directory tree. Crawler, indexer and search frontend are all part of the same system, and crawls can be scheduled or run in realtime.

Running it is a server job rather than a laptop one: the documented setup targets x86-64 Linux with Docker, JDK 25 and liburing, and a production-like index wants roughly 32 GB of RAM or more, fast SSDs for the index and additional terabytes of slower disks for crawl data. It is not a substitute for the index size or freshness of Google or Bing, and the project does not aim to be — it is an alternative for people who want to discover non-commercial content, or who want their own search backend that they fully control.

Frequently Asked Questions

What are the system requirements to self-host Marginalia Search?

The documented setup targets x86-64 Linux with Docker, JDK 25 and liburing, since the index uses io_uring. A production-like environment needs a lot of RAM, about 32 GB as a floor, plus fast SSDs for the index and additional terabytes of slower disks for crawl data. A local developer deployment can run with a much smaller index.

Can I index data I already have instead of crawling the web?

Yes. Besides crawling, the engine supports sideloading existing content in several formats: WARC files, ZIM archives, StackExchange dumps and a plain directory tree. This is the usual path for running it as a white-label search engine over your own corpus.

Is Marginalia Search suitable as a document search engine for a single website or intranet?

Not really. The documentation explicitly contrasts it with general document search engines like Elasticsearch or Meilisearch: Marginalia makes assumptions about internet data and is described as relatively poorly suited for use as a document search engine. Use it when you want to search a crawl of many sites, not a document collection.

Does Marginalia Search include its own crawler?

Yes, crawling is integrated into the same system as the indexer and the search frontend. Crawls can be scheduled and re-run, and the docs cover realtime crawling as well as exporting crawl data back out as WARC files.

Looking for contributors

This project is actively seeking help, join the community!

Visit Marginalia Search