Node.js

Supports vision models from different providers like OpenAI, Azure OpenAI, Anthropic, AWS Bedrock, Google Gemini, etc.

​

Installation

npm install zerox

Zerox uses graphicsmagick and ghostscript for the PDF to image processing step. These should be pulled automatically, but you may need to manually install.

sudo apt-get update
sudo apt-get install graphicsmagick ghostscript

​

Usage

​

With file URL

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,
  },
});

​

From local path

import { zerox } from 'zerox';
import path from 'path';

const result = await zerox({
  filePath: path.resolve(__dirname, '<path-to-your-file>'),
  credentials: {
    apiKey: process.env.OPENAI_API_KEY,
  },
});

​

Params

The maintainFormat option tries to return the markdown in a consistent format by passing the output of a prior page in as additional context for the next page. This requires the requests to run synchronously, so it’s a lot slower. But valuable if your documents have a lot of tabular data, or frequently have tables that cross pages.

​

Supported Models

Zerox supports a wide range of models across different providers:

{
  apiKey, // Required
}

{
  apiKey, // Required
  endpoint, // Required
}

{
  accessKeyId, // Required if no sessionToken
  region, // Required if no sessionToken
  secretAccessKey,
  sessionToken, // Required if no accessKeyId or secretAccessKey
}

{
  apiKey, // Required
}

​

Data Extraction

Zerox supports structured data extraction from documents using a schema. This allows you to pull specific information from documents in a structured format instead of getting the full markdown conversion.

Set extractOnly: true and provide a schema to extract structured data. The schema follows the JSON Schema standard.

Use extractPerPage to extract data per page instead of from the whole document at once.

You can also set extractionModel, extractionModelProvider, and extractionCredentials to use a different model for extraction than OCR. By default, the same model is used.

​

Response

Example

{
  "completionTime": 10038,
  "fileName": "invoice_36258",
  "inputTokens": 25543,
  "outputTokens": 210,
  "pages": [
    {
      "page": 1,
      "content": "# INVOICE 36258 ...",
      "contentLength": 747
    }
  ],
  "extracted": null,
  "summary": {
    "totalPages": 1,
    "ocr": {
      "failed": 0,
      "successful": 1
    },
    "extracted": null
  }
}