Aggregating web content into a consolidated, LLM-ready context.
llm-txt-resolver is a tool for crawling an llms.txt, html page, or markdown document and aggregating connected links into a single clean text block.
- Recursive Crawling: Traverses a document by following links, up to a configurable depth.
- Content Extraction: Intelligently extracts the main content from html and markdown pages.
- Caching: Caches content to speed up subsequent runs and reduce network requests.
- Knowledge Graph: Builds a graph of the site's structure, which can be used for further analysis.
- Node & Browser: Works both in Node.js and in browser. Uses file system for Node.js caching and
CacheAPI for browser caching.
npm i llm-txt-resolver
# or bun
bun i llm-txt-resolvernpx llm-txt-resolver https://firebase.google.com/docs/ai-logic/llms.txt ai/llms.txt --depth 2 # default is 2import { Resolver } from 'llm-txt-resolver'
const resolver = new Resolver()
const llmsTxt = 'https://firebase.google.com/docs/ai-logic/llms.txt'
// Promised based
const { content, graph } = await resolver.resolve(llmsTxt)
console.log({ content, graph })
// Stream based
const stream = resolver.resolve(llmsTxt, { stream: true })
for await(let chunk of stream) {
console.log(chunk)
}The browser build of llm-txt-resolver is configured to use the BrowserCacheProvider by default, which leverages the browser's Cache API for caching. This makes it easy to use the resolver in a browser environment without any additional configuration.
To use the browser version, import from llm-txt-resolver/browser:
import { Resolver } from 'llm-txt-resolver/browser'
const resolver = new Resolver()
const llmsTxt = 'https://firebase.google.com/docs/ai-logic/llms.txt'
const { content, graph } = await resolver.resolve(llmsTxt)
console.log({ content, graph })
const stream = resolver.resolve(llmsTxt, { stream: true })
for await(let chunk of stream) {
console.log(chunk)
}The Resolver in the browser build has the same API as the Node.js version, but with BrowserCacheProvider as the default cache provider. You can still provide your own cacheProvider in the constructor if you wish to override the default behavior.
The main class for resolving web content.
Creates a new Resolver instance.
options(optional): An object with the following properties:depth(optional, default:2): The maximum depth to crawl.cacheProvider(optional, default:new FileSystemCacheProvider()in Node.js,new BrowserCacheProvider()in the browser): An instance of a class that implements theCacheProviderinterface.
resolve(rootUrl: string, options?: ResolveOptions): Promise<{ content: string; graph: KnowledgeGraph }>
Starts the crawling process from the given rootUrl.
rootUrl: The URL to start crawling from.options: An object with the following properties:stream(optional, default:false): Whether to stream the output.
- Returns a promise that resolves to an object with the following properties:
content: A single string containing the aggregated content of the website.graph: AKnowledgeGraphinstance representing the structure of the site.
An interface for implementing custom caching strategies.
export interface CacheProvider {
load(rootUrl: string): Promise<KnowledgeGraph | null>;
save(rootUrl: string, graph: KnowledgeGraph): Promise<void>;
}The default Node.js cache provider, which stores the cache on the local file system.
cacheDir(optional, default:.cache): The directory to store the cache in.
The default Browser cache provider, which uses the Cache API to store the cache.
Contributions are welcome! Please open an issue or submit a pull request.