smallweb-search
Marginalia Search
An independent search engine with its own crawler, focused on the small, old, text-heavy, non-commercial web. You can search on the site, and its open public API lets this site query it directly. Reach for it when hunting personal homepages, hobby sites, and long-form text from the web's quieter corners that mainstream engines tend to bury.
Why it’s useful & how it works
Marginalia is the engine to try when mainstream results are dominated by commercial sites and you are specifically looking for personal homepages, long-form technical writing, or hobby pages. It runs its own crawler, deliberately scores down ad-heavy and SEO-optimised sites, and surfaces the text-heavy, non-commercial corners of the web that larger engines tend to bury. Echo can query Marginalia directly through its public API, so results appear alongside those from other sources without leaving the page.
What’s inside
Marginalia's index holds approximately 300 to 400 million documents as of 2024, with the creator actively working toward a billion-document goal. The index is smaller than those of mainstream engines by design, prioritising quality and independence over raw size.
API access
https://api.marginalia.nu/public/search/ <q>?key=public
What we measured
Our own probes, not the archive’s own claims. Re-run periodically; every reading below is dated.
Reachability
- Direct request
- Responded HTTP 200 537 ms
- Through a datacenter proxy
- Responded HTTP 200 979 ms
- API, direct
- Responded HTTP 200 420 ms
- API, through a proxy
- Responded HTTP 200 1.6 s
Did a real search return anything?
Returned results · 20 results for the sample query · median 355 ms · stable across runs
Sample query: https://en.wikipedia.org/wiki/Coffee
Benchmark coverage
Found 7 of 10 test items (70%) in the Web full-text cluster. Median response 325 ms.
By how well known the item is: 3/3 well-known, 2/4 mid, 2/3 obscure .
One item it found: https://en.wikipedia.org/wiki/Coffee
Reachability measured 2026-08-22. Search checked 2026-08-22. Benchmark run 2026-08-22.
Access
Programmatic API access (a key may be required, see the API tag).