Docling Core

TypeScript definitions and functions for using Docling output, which simplifies document processing, parsing diverse formats — including advanced PDF understanding. This package supports integrating Docling output into your Type-/JavaScript app with type definitions and utility functions that reflect the Docling Core Python library.

Getting started

Install package

npm i @docling/docling-core

Convert and fetch

To convert a document you can:

• Use a separate document conversion service, such as docling-serve.
• Integrate Docling into your own API, or use the Docling CLI for a quick web service.
• Pre-convert documents and host these as static resources, on a COS bucket for example.

Next, fetch the (typed) conversion:

import { type DoclingDocument } from '@docling/docling-core';

async function fetchConversion(url: string) {
  const response = await fetch(url);

  return (await response.json()) as DoclingDocument;
}

Iterate and discriminate

Use the utility and typing functions for easy access to the converted document contents:

import { iterateDocumentItems, isDocling } from '@docling/docling-core';

for (const [item, level] of iterateDocumentItems(conversion)) {
  if (isDocling.TextItem(item)) {
    console.log(item.text);
  } else if (isDocling.TableItem(item)) {
    ...
  }
}

Read DocLang

@docling/docling-core/doclang turns DocLang XML into a DoclingDocument. It is a port of the Python DocLangDocDeserializer, and its JSON output matches Python's.

import { deserializeDocLang } from '@docling/docling-core/doclang';

const doc = deserializeDocLang(doclangXml); // browser: uses the native DOMParser

In Node, pass a DOMParser. @xmldom/xmldom is an optional peer dependency, so browser bundles never include it.

import { DOMParser } from '@xmldom/xmldom';

const doc = deserializeDocLang(doclangXml, { DOMParser });

The options mirror the Python deserialize_str keyword arguments:

• pageNo: the page of content before the first <page_break/> (default 1).
• maxXmlBytes, maxDepth, maxElements: budgets for untrusted input. Exceeding one throws.
• repairUnescapedText: escape a bare &, a < that cannot start markup, a stray ]]> and illegal control characters before parsing. Off by default, so the input must be well-formed XML.
• onWarning: receives warnings (default console.warn). <track> is not supported yet and is skipped, with all its content, with a warning.

Read DCLX archives

deserializeDclx reads a DCLX archive from bytes (Uint8Array, ArrayBuffer, Blob or File). Images are embedded in the document as base64 data URIs.

import { deserializeDclx } from '@docling/docling-core/doclang';

const doc = await deserializeDclx(await (await fetch(url)).arrayBuffer());

In Node, pass a DOMParser like with DocLang.

Limitations

This package does not

• Convert documents directly from within a JavaScript runtime. It is not a port of Docling.
• Support Docling output formats other than JSON and DocLang (read-only), such as Markdown.