Marginalia Search review: the noncommercial web

Marginalia Search is a search engine for the noncommercial web. It is for people who want text-heavy personal sites, and the project says your internet address cannot historically be tied to your searches.¹² The project crawls the web itself.⁸ It prefers pages with a lot of writing, and it is still run by one person.¹

Last checked October 28, 2026.

At a glance

What it is: A search engine that favors text-heavy, non-commercial websites.¹

Works on: A web browser. An older layout is at old-search.marginalia.nu.⁸

Price: Search the public website in a browser. For programs, a test key called public is free and often rate-limited, a non-commercial key is free by email, and paid keys are optional.³ Donations and API-key sales were rising in October 2026.⁵

Licence: Most source code is AGPL 3.0. Some parts are also under MIT. Non-commercial API use is CC-BY-NC-SA 4.0.⁴³

Best for: Finding personal, text-heavy sites.

Official site: marginalia-search.com

What Marginalia Search does (and how it protects your privacy)

A crawler is a program that visits public pages and stores them in an index, a searchable copy. Marginalia Search runs its own crawler, which the site says identifies itself as search.marginalia.nu and respects robots.txt, the file a site uses to say what may be fetched.⁸ The about page says the index holds hundreds of millions of documents.¹ On 6 October 2025 the developer wrote that the total had reached 969 million.⁶

The about page says the engine uses heuristics, simple rules of thumb, to prefer text-heavy websites, and puts an upper limit on heavy modern design.¹ The aim is the non-commercial internet, and most of those links are nearly impossible to find on a regular search engine.¹ It calls itself a niche tool, and the wrong choice when you need facts you can trust.¹ What your internet provider can still see is separate from what this engine stores.

Key features of Marginalia Search

Text-heavy pages are preferred, and heavy modern design is limited.¹
The crawler uses the name search.marginalia.nu and follows robots.txt.⁸
Quotes match an exact phrase, in that order, with no words in between.⁸
A minus sign drops a word, as in soup -chicken.⁸
site: searches one domain. browse: lists similar sites.⁸
-special:tracking hides pages marked as carrying analytics or tracking code. A coffee-stain mark shows where your words sit on the page.⁸
Custom filters can boost chosen sites or words. They stay in local storage, a small save area in your browser, and are sent only when you use them.⁷

Pros and cons

Pros

It is built to surface non-commercial pages that a large engine rarely ranks.¹
The privacy page says clicks are not recorded, and that your internet address cannot historically be tied to your searches.²
The source is public under AGPL 3.0, so the way the engine works can be read.⁴
On 26 October 2026 the developer said recent work was mostly bug fixes, so the project is still maintained.⁵

Cons

If you want facts you can trust, the project says this is almost certainly the wrong tool.¹
English is the language the index grew from. German, French, and Swedish were an October 2025 experiment with a very small set of pages.⁶
One person runs it, and the October 2026 update says it is not yet perpetually self-sustaining.¹⁵
Running your own copy needs a lot of memory. At 32 GB it may still not perform very well.⁴
The shared public API key often hits a rate limit, a cap on how often it can be called, and cannot create custom filters.³

Is Marginalia Search safe and private?

The privacy page, dated 22 September 2022, says logging stays minimal because queries are sensitive.² It does not record which links you click, and it says an IP address, the number that identifies your connection, cannot historically be matched to search terms.² A short window of logs is kept for rate limits and to diagnose bugs, then logs that may contain queries are shredded.² It also says enough time and a large forensics budget could probably still work out who you are and what you searched, and that measures are taken to make that time consuming and expensive.²

On 26 October 2026 the developer wrote that there are no analytics.⁵ Custom filters stay in the browser and are sent only when you use them, rather than always providing an identification cookie, a small file a site can use to recognize you later.⁷ One person operates the engine.¹ The code is mostly AGPL 3.0, with some parts also under MIT, and the README records funding through NLnet’s NGI0 Entrust Fund, with support from the European Commission’s Next Generation Internet programme.⁴ The October 2026 note adds a $10,000 SerpApi sponsorship and says the operation is not yet perpetually self-sustaining.⁵ Non-commercial API results use CC-BY-NC-SA 4.0. Commercial keys are metered and do not require attribution.³

How to get started with Marginalia Search

Open marginalia-search.com. That page is the search box. API keys are for a program that calls the service.³
Type a few words. For an exact phrase, use quotes. The help text says “steve mcqueen” matches those words in that order, with no words in between.⁸
Put a minus before a word you do not want. soup -chicken looks for keywords that contain soup, but not chicken.⁸
Use site: and a domain to stay on one website, or browse: and a domain to see similar sites. Both are in the on-site help.⁸
Add -special:tracking to hide pages marked as having analytics or tracking code. -special:affiliate does the same for likely Amazon affiliate links.⁸
If the new layout is awkward, open old-search.marginalia.nu. The site says it is functionally nearly identical.⁸
For a script of your own, read the API page. The public key is there for a test. A free non-commercial key means emailing [email protected], which the October 2026 update says can take a week or two.³⁵

Who should use Marginalia Search (and who might prefer something else)

Use it for a personal site or a long explanation. The about page says it helps you find things you did not know you were looking for, and that it is not the greatest at finding what you already knew was there.¹ Keep a large general engine for news and shopping. Save a page for later with wallabag.

To run an index on a computer you control, see YaCy. Marginalia Search can run as your own copy too, but that needs a lot of memory, and 32 GB may still not perform very well.⁴ For a nonprofit metasearch, see MetaGer.

Frequently asked questions

Is Marginalia Search free?

Search the public website in your browser. For programs, there is a public test key, a free non-commercial key by email under CC-BY-NC-SA 4.0, and optional paid keys.³ The README says non-commercial share-alike use is always free.⁴ The October 2026 note says the operation is not yet perpetually self-sustaining.⁵

Is Marginalia Search safe?

The privacy page says logging is minimal, clicks are not stored, and an internet address cannot historically be matched to queries. Brief query logs are then shredded. It also says a large investigation could probably still reconstruct a search.² In October 2026 the developer wrote that the project has no analytics.⁵

Does Marginalia Search work on iPhone?

Yes. Open marginalia-search.com in Safari or any other browser. For fewer visual tricks, old-search.marginalia.nu is described as functionally nearly identical.⁸

Does Marginalia Search only work in English?

English is the language the index grew from, by following links on English-language sites.⁶ On 6 October 2025, experimental German, French, and Swedish search went live, with what the developer called an extremely small corpus, a very small pile of pages.⁶