Mwmbl

Last reviewed Sep 13, 2026 · 797 words

Mwmbl is not a SearXNG replacement, and if you install it expecting one you will be disappointed inside an hour. SearXNG forwards your query to Google, Bing and dozens of other engines; Mwmbl has its own index, built by volunteers, and that index is still small enough that a query for anything local, recent or obscure often returns nothing useful. What Mwmbl offers is different and, I think, more interesting: it is the only web search project I know of where one person can run the whole thing, crawler to results page, on a 1 GB box, and where ranking is curated by people rather than by an ad auction.

The index is the product, and it is still small

Mwmbl is run as a non-profit, licensed AGPL-3.0, written in Python, with 1,852 GitHub stars at the time of writing and a first release in 2021. The index is built by crowd-sourced crawling: volunteers run a browser extension or a standalone crawler that fetches pages, extracts terms and pushes the result to the central service. The project then publishes the index so anyone can serve it.

The consequence is that coverage follows where volunteers have crawled. Programming questions, Wikipedia-adjacent topics and well-linked English pages are reasonably covered. A plumber in your town or a news story from yesterday is not. I treat it as a curated reference index rather than a general search engine, and judged on those terms it holds up.

Self-hosting means serving a downloaded index, not building your own

A self-hosted Mwmbl has three parts: the Django-based web app and API, the index files, and optionally a crawler that contributes back. The realistic self-host is the first two. Building an independent index from scratch would need crawl volume no homelab can supply, so you fetch the published index and serve it locally. That gives you a search page that answers with zero outbound requests, which is a genuine privacy property: no upstream engine sees your query, because there is no upstream engine.

Deployment is Docker or bare metal and the difficulty is Medium, mostly because the documentation assumes you have read the code. Budget 1 GB of RAM as a floor and put the index on an SSD, since more page cache means fewer disk reads per query. The hard part is not the install but keeping your copy in step with the published index, which changes as volunteers crawl and curate. Plan on a scripted refresh, weekly at most, and accept that your instance is always a little behind mwmbl.org.

Curation is the feature nobody else has

Logged-in users on mwmbl.org can drag results into a better order for a query, and those edits feed the ranking. It is a Wikipedia-style bet: a few thousand motivated people produce better results for common queries than an opaque ranking model does. On a self-hosted instance you serve the curated index as of your last refresh, but you do not receive ongoing edits between refreshes. If your reason for getting involved is to contribute, run the crawler rather than the search page. The crawler is what the project actually needs; another copy of the results page is not.

Where it sits next to SearXNG and YaCy

For daily search, run SearXNG. It is the safer pick by a wide margin, with a far larger community and results as good as the engines it aggregates, and it is the default I recommend to anyone browsing the search category. YaCy is the other self-crawled option: a peer-to-peer index in Java that has been around since the mid-2000s, heavier (plan on 2 GB or more) and noisier in its results, but every YaCy peer really does crawl. Mwmbl sits between them, lighter than YaCy and more curated, with a central non-profit coordinating the work.

What I'd do

Keep SearXNG as the browser default. Run the Mwmbl crawler on a spare box or in the browser extension if the project's goal matters to you, and put a bookmark to mwmbl.org next to it. Self-host the Mwmbl search page in only two situations: an air-gapped or heavily filtered network where an offline reference index is worth 1 GB of RAM, or a classroom or workshop where showing people a search engine whose source they can read is the point. For everyone else the honest answer in 2026 is contribute crawling, search with SearXNG, and check back in a year as the index grows.