Zerox

// Overview

Zerox is an OCR tool that turns documents into Markdown for AI ingestion. Instead of relying on traditional text-extraction rules, it converts each page of a file (PDF, DOCX, image, and more) into an image, sends that image to a vision model, asks for Markdown, and aggregates the per-page responses into one result.

It is aimed at developers who need to feed documents with awkward layouts, tables, and charts into an LLM pipeline. Because vision models read a page the way a person would, Zerox tends to hold up on content that line-based parsers struggle with.

Within the RAG and retrieval space, Zerox fits the document parsing and ingestion step: it produces the clean Markdown you then chunk, embed, and index. It ships as both a Node package (zerox on npm) and a Python package (py-zerox on PyPI), and supports OpenAI, Azure OpenAI, AWS Bedrock, and Google Gemini, with Vertex AI on the Python side.

// What it does

• Converts PDF, DOCX, and image files into Markdown by sending page images to a vision model
• Works with multiple providers: OpenAI, Azure OpenAI, AWS Bedrock, and Google Gemini (plus Vertex AI in Python)
• maintainFormat option carries a prior page's output into the next request to keep tables and formatting consistent across pages
• Concurrent page processing via the concurrency option, with configurable retries and error modes
• Structured data extraction with a schema, including per-page extraction (Node)
• Image controls like orientation correction, edge trimming, DPI/imageDensity, and page selection

// Getting started

Zerox is available as both a Node and a Python package. The Node quickstart below OCRs a PDF and returns Markdown.

npm install zerox

import { zerox } from "zerox";

const result = await zerox({
  filePath: "https://omni-demo-data.s3.amazonaws.com/test/cs101.pdf",
  credentials: {
    apiKey: process.env.OPENAI_API_KEY,
  },
});

const result = await zerox({
  filePath: "path/to/file.pdf",
  credentials: { apiKey: process.env.OPENAI_API_KEY },
  maintainFormat: true,
  concurrency: 10,
});

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Ingesting scanned or image-heavy PDFs into a RAG pipeline as clean Markdown for chunking and embedding
• Extracting tables and structured data from invoices, receipts, or reports using a schema
• Parsing documents with awkward layouts, charts, or multi-column text that line-based OCR mishandles
• Batch-converting a mix of PDFs, DOCX files, and images into Markdown for an LLM workflow

// How Zerox compares

Zerox alongside other open-source parsing & ingestion tools AI/TLDR tracks, ranked by GitHub stars.