Recursive URL Loader

When loading content from a website, we may want to process load all URLs on a page.

For example, let's look at the LangChain.js introduction docs.

This has many interesting child pages that we may want to load, split, and later retrieve in bulk.

The challenge is traversing the tree of child pages and assembling a list!

We do this using the RecursiveUrlLoader.

This also gives us the flexibility to exclude some children, customize the extractor, and more.

Setup

To get started, you'll need to install the jsdom package:

npm

npm i jsdom

Yarn

yarn add jsdom

pnpm

pnpm add jsdom

We also suggest adding a package like html-to-text or @mozilla/readability for extracting the raw text from the page.

npm

npm i html-to-text

Yarn

yarn add html-to-text

pnpm

pnpm add html-to-text

Usage

import { compile } from "html-to-text";
import { RecursiveUrlLoader } from "@langchain/community/document_loaders/web/recursive_url";

const url = "/docs/introduction";

const compiledConvert = compile({ wordwrap: 130 }); // returns (text: string) => string;

const loader = new RecursiveUrlLoader(url, {
  extractor: compiledConvert,
  maxDepth: 1,
  excludeDirs: ["/docs/api/"],
});

const docs = await loader.load();

Options

interface Options {
  excludeDirs?: string[]; // webpage directories to exclude.
  extractor?: (text: string) => string; // a function to extract the text of the document from the webpage, by default it returns the page as it is. It is recommended to use tools like html-to-text to extract the text. By default, it just returns the page as it is.
  maxDepth?: number; // the maximum depth to crawl. By default, it is set to 2. If you need to crawl the whole website, set it to a number that is large enough would simply do the job.
  timeout?: number; // the timeout for each request, in the unit of seconds. By default, it is set to 10000 (10 seconds).
  preventOutside?: boolean; // whether to prevent crawling outside the root url. By default, it is set to true.
  callerOptions?: AsyncCallerConstructorParams; // the options to call the AsyncCaller for example setting max concurrency (default is 64)
}

However, since it's hard to perform a perfect filter, you may still see some irrelevant results in the results. You can perform a filter on the returned documents by yourself, if it's needed. Most of the time, the returned results are good enough.