LlamaIndex adapter

@nalinor/mupdf4llm ships a LlamaIndex reader at the @nalinor/mupdf4llm/llama subpath export — a one-for-one port of pymupdf4llm.llama.PDFMarkdownReader.

Setup

llamaindex is declared as an optional peer dependency. Install it in your project if you want a real Document:

npm i llamaindex

If you don't, the reader still works — it returns plain { text, metadata } records with the same field names as Document, so consumer code can read .text and .metadata without branching on whether llamaindex is installed.

Usage

import { PDFMarkdownReader } from "@nalinor/mupdf4llm/llama";

const reader = new PDFMarkdownReader();
const docs = await reader.loadData("paper.pdf");
// docs[i] is a Document (or { text, extra_info }) — one per page

Filter metadata

const reader = new PDFMarkdownReader({
  metaFilter: (m) => ({ page: m.page, title: m.title }),
});

Forwarding MarkdownOptions

loadData(path, extra?, opts?) forwards a third argument to toMarkdown — handy if you want images, words, etc. inside each Document:

const docs = await reader.loadData(
  "paper.pdf",
  {},
  {
    embedImages: true,
    extractWords: true,
  },
);

End-to-end example

See examples/llama-rag for a complete RAG pipeline using VectorStoreIndex + an OpenAI / Anthropic-backed LLM.