HTML in the sandbox

Specifier: @nodetool-ai/sandbox-html. Import it at the top of the body.

cheerio imports 25 Node builtins and turndown wants a DOM, so neither compiles into the guest. This pack is a host module: the import resolves to a generated facade over NodeTool's own implementation. Both exports are async.

select — CSS selection

import { select } from "@nodetool-ai/sandbox-html";

const hrefs = await select(inputs.html, "a[href]", { attr: "href", limit: 500 });
const titles = await select(inputs.html, "h2");   // trimmed text of each match
return { hrefs, titles };

Returns trimmed text per match, or the named attribute when attr is set. limit defaults to 100 and is clamped to 1000 — the cap is in the host implementation, so raising it in the call does nothing.

toMarkdown — a whole page

import { toMarkdown } from "@nodetool-ai/sandbox-html";

const markdown = await toMarkdown(inputs.html);
return { markdown };

The usual first step before summarizing or indexing a fetched page. Pass { turndown: {...} } to override turndown's own options.

toText — a whole page as plain text

import { toText } from "@nodetool-ai/sandbox-html";

const text = await toText(inputs.html);
return { text };

Strips <script>/<style> and every tag, then collapses whitespace.

extractLinks / extractImages / extractAudio / extractVideos

import {
  extractLinks,
  extractImages,
  extractAudio,
  extractVideos
} from "@nodetool-ai/sandbox-html";

const links = await extractLinks(inputs.html, inputs.baseUrl);
// [{ href, text, type: "internal" | "external" }, ...]

const images = await extractImages(inputs.html, inputs.baseUrl); // string[] of resolved URLs
const audio = await extractAudio(inputs.html, inputs.baseUrl);
const videos = await extractVideos(inputs.html, inputs.baseUrl);

baseUrl is optional. When set, relative src/href values resolve against it and extractLinks classifies same-origin links as internal. Each extractor returns at most 1000 items.

extractMetadata — title, description, keywords

import { extractMetadata } from "@nodetool-ai/sandbox-html";

const { title, description, keywords } = await extractMetadata(inputs.html);

Any field the page doesn't set comes back null.

extractReadableText — strip chrome, keep the article

import { extractReadableText } from "@nodetool-ai/sandbox-html";

const article = await extractReadableText(inputs.html);

Removes script/style/nav/aside/footer/header, then returns the text of the first match among article, main, [id*=content], [class*=content], or body. A page with none of those returns "No main content found".

Getting a page's base URL

No import needed — new URL(pageUrl).origin gives the base URL to pass into the extractors above.

Gotchas

• Every export is async.
• 5 MB of HTML per call. Larger input is refused by name.
• A selector that matches nothing returns [], never null. An invalid selector throws with cheerio's own message.
• Extractors cap at 1000 items. Anything past that is silently dropped.