Plumb

Plumb Search

A free, open-source search engine for sites, articles, places and more. Search here, or run it on your own computer so your searches stay with you.

This box searches the Plumb node on this server.

Open sourceNo adsYour own index

What it finds

Websites

Site names, sitelinks and a learned ranking model that tries to put the official site first.

Try Rust programming →

Reference

Wikipedia, books, papers, music, films and podcasts alongside web results.

Try Marie Curie →

Fully hackable

All of Plumb is open source, written in Rust, under the MIT or Apache-2.0 license, and every copy you run is yours to change. Add results from any source with a plugin: a small Rust program the node runs in a sandbox, which can reach only the addresses it lists. Read results as JSON, change how it ranks, or fork the whole thing.

Write a plugin · Source code

Web search for AI apps

No API key, no quota, and Plumb doesn't scrape Google, Bing or anyone else: it answers from its own index, so nothing upstream can block or throttle it. It speaks MCP and SearXNG's JSON API, so it drops into the setup you already have.

claude mcp add --transport http plumb https://plumbsearch.org/mcp

Apps that take a SearXNG address (Open WebUI, LibreChat, AnythingLLM and others) can use https://plumbsearch.org, which answers /search?q=<query>&format=json.

It is good at what agents look up most: official sites, packages and their docs, well-known Stack Overflow questions, Wikipedia facts and instant answers. It indexes homepages and those page sets, not the full text of the web, so the long tail is weaker than Google's. Each client gets a burst of 30 MCP tool calls, then 60 a minute.

Setup for LM Studio, Open WebUI, Jan and more · MCP tools

Search here, or run your own

Search right here on plumbsearch.org. If you'd rather keep your searches to yourself, run your own copy on a desktop or home server; it's optional.

More than names

For each site Plumb keeps its name, homepage title, description and key terms, what other sites call it, and its kind and country from Wikidata. It also matches by meaning, lists Wikipedia articles, Stack Overflow questions, packages, books, papers, podcasts and places, and orders results with a small learned ranking model. So “banks” lists real banks, and “electric car maker” finds carmakers.

Search across the Plumb network

Plumb nodes share crawls and can ask each other for results. Desktop and Docker join the network by default. Use the search settings to include network results.

Private and meaning search

Search privately in your browser so your query stays on your device, or find sites by topic with meaning search. Private search uses WebAssembly in your browser; ordinary search works without JavaScript.

Try private search

Get Plumb

Desktop app

For Windows, macOS (Apple silicon and Intel) and Linux (.deb, .rpm and AppImage). It shows the search page in its own window and opens results in your usual browser.

Download the latest release

The installers aren't code-signed yet, so your computer will warn you the first time you open the app.

Server or homelab

One container for Linux on x86 or ARM, including a 64-bit Raspberry Pi. The search page is on port 8080, and port 4001 is for the Plumb network.

docker run -d --name plumb --init \
  --restart unless-stopped --stop-timeout 300 \
  -p 8080:8080 -p 4001:4001/tcp -p 4001:4001/udp \
  -v plumb-data:/data \
  ghcr.io/sueheir/plumb-search:latest

:latest follows each new release; :0.2 stays on this one.

Docker guide: settings, updates and backups.

Desktop app

For Windows, macOS (Apple silicon and Intel) and Linux (.deb, .rpm and AppImage). Its window controls your node; searches and results open in your usual browser.

Build the desktop app

Installers are not published yet. Check future releases, or use Docker today.

Make it your browser's search engine. Open your Plumb page in Firefox, right-click the address bar and choose “Add Plumb Search”. Other browsers can add it by hand with the search address https://plumbsearch.org/search?q=%s for this site, http://127.0.0.1:7586/search?q=%s for the desktop app, or your server's address followed by /search?q=%s.

For site owners

Plumb visits homepages to fill in each site's title and to find new sites through its links. Its crawler:

Two other kinds of request come from Plumb. When someone's AI assistant asks their own node to read a page, the node fetches that one page as plumb-mcp/<version> (+https://github.com/SueHeir/plumb-search). And each node downloads the European Central Bank's daily exchange rates for currency conversions, and Open-Meteo's forecast for a town when someone asks its weather.

Every copy of Plumb does its own crawling, so visits can come from many addresses, all following the same rules. To keep Plumb out, add this to your robots.txt:

User-agent: PlumbSearch
Disallow: /

Questions or problems? Open an issue on GitHub.

Where it's going

Copies of Plumb share their index with each other and take turns checking sites in the background. Next, they will agree on what's popular without revealing who searched for what. No accounts and no ads.

  1. One machine. Build an index from public data and put the official site first.
  2. Shared index. Copies stay in sync with each other, with no central server needed.
  3. Community crawling. Copies take turns checking sites and sign what they crawl, so the index stays fresh on its own.
  4. Private popularity. Count what people pick without tracking anyone, and keep bots from gaming it. It works today but is off unless a node turns it on (--share-popularity, or on the settings panel).