Documind

Documind is an advanced document processing tool that leverages AI to extract structured data from PDFs. It is built to handle PDF conversions, extract relevant information, and format results as specified by customizable schemas.

Features

• Extracts structured JSON output from unstructured documents.
• Converts documents into Markdown format.
• Supports custom schemas for data extraction.
• Includes pre-defined templates for common schemas.
• Works with OpenAI and custom LLM setups (Llava and Llama3.2-vision).
• Auto-generates schemas based on document content.

Try the Hosted Version 🚀

The hosted version provides a seamless experience with fully managed APIs, so you can skip the setup and start extracting data right away. Join the beta to get access to the hosted service.

Roadmap

✅ Released Features

• [x] PDF Extraction
• [x] Basic Schema Definition
• [x] Structured JSON Output
• [x] Template Schemas
• [x] Local LLM Integration (Llama3.2)
• [x] Auto-generated Schemas
• [x] Document Formatters (Text and Markdown)
• [x] Multi-file Support (DOCX, PNG, JPG, TXT, HTML)
• [x] Additional Schema Field Types (Boolean and Enum)

🚧 Upcoming Features

• [ ] Additional Models (Local and cloud)
• [ ] Image Extraction
• [ ] Advanced Document Formatters
• [ ] Data Classification
• [ ] Use Your Fine-tuned Models

Requirements

Before using documind, ensure the following software dependencies are installed:

System Dependencies

• Ghostscript: documind relies on Ghostscript for handling certain PDF operations.
• GraphicsMagick: Required for image processing within document conversions.

Install both on your system before proceeding:

# On macOS
brew install ghostscript graphicsmagick

# On Debian/Ubuntu
sudo apt-get update
sudo apt-get install -y ghostscript graphicsmagick

Node.js & NPM

Ensure Node.js (v18+) and NPM are installed on your system.

Installation

You can install documind via npm:

npm install documind

Environment Setup

documind requires an .env file to store sensitive information like your OpenAI API key.

Create an .env file in your project directory and add the following:

OPENAI_API_KEY=your_openai_api_key

Usage

Basic Example

First, import documind and define your schema. The schema outline what information documind should look for in each document. Here’s a quick setup to get started.

1. Define a Schema

The schema is an array of objects where each object defines:

• name: Field name to extract.
• type: Data type (e.g., "string", "number", "array", "object").
• description: Description of the field.
• children (optional): For arrays and objects, define nested fields.

Example schema for a bank statement:

const schema = [
  {
    name: "accountNumber",
    type: "string",
    description: "The account number of the bank statement."
  },
  {
    name: "openingBalance",
    type: "number",
    description: "The opening balance of the account."
  },
  {
    name: "transactions",
    type: "array",
    description: "List of transactions in the account.",
    children: [
      {
        name: "date",
        type: "string",
        description: "Transaction date."
      },
      {
        name: "creditAmount",
        type: "number",
        description: "Credit Amount of the transaction."
      },
      {
        name: "debitAmount",
        type: "number",
        description: "Debit Amount of the transaction."
      },
      {
        name: "description",
        type: "string",
        description: "Transaction description."
      }
    ]
  },
  {
    name: "closingBalance",
    type: "number",
    description: "The closing balance of the account."
  }
];

2. Run documind

Use documind to process a PDF by passing the file URL and the schema.

import { extract } from 'documind';

const runExtraction = async () => {
  const result = await extract({
    file: 'https://bank_statement.pdf',
    schema
  });

  console.log("Extracted Data:", result);
};

runExtraction();

Example Output

Here’s an example of what the extracted result might look like:

 {
  "success": true,
  "pages": 1,
  "data": {
    "accountNumber": "100002345",
    "openingBalance": 3200,
    "transactions": [
        {
        "date": "2021-05-12",
        "creditAmount": null,
        "debitAmount": 100,
        "description": "transfer to Tom" 
      },
      {
        "date": "2021-05-12",
        "creditAmount": 50,
        "debitAmount": null,
        "description": "For lunch the other day"
      },
      {
        "date": "2021-05-13",
        "creditAmount": 20,
        "debitAmount": null,
        "description": "Refund for voucher"
      },
      {
        "date": "2021-05-13",
        "creditAmount": null,
        "debitAmount": 750,
        "description": "May's rent"
      }
    ],
    "closingBalance": 2420
  },
  "fileName": "bank_statement.pdf"
}

Read the documentation for more on how to define schemas and and enable auto-generation.

Templates

Documind comes with built-in templates for extracting data from popular document types like invoices, bank statements, and more. These templates make it easier to get started without defining your own schema.

List available templates

You can list all available templates using the templates.list function.

import { templates } from 'documind';

const templates = templates.list();
console.log(templates); // Logs all available template names

Use a template

To use a template, simply pass its name to the extract function along with the file you want to extract data from. Here's an example:

import { extract } from 'documind';

const runExtraction = async () => {
  const result = await extract({
    file: 'https://bank_statement.pdf',
    template: 'bank_statement'
  });

  console.log("Extracted Data:", result);
};

runExtraction();

Read the templates documentation for more details on templates and how to contribute yours.

Using Local LLM Models

Read more on how to use local models here.

Contributing

Contributions are welcome! Please submit a pull request with any improvements or features.

License

This project is licensed under the AGPL v3.0 License.

Credit

This repo was built on top of Zerox. The MIT license from Zerox is included in the core folder and is also mentioned in the root license file.