Intelligent document processing using State of the Art AI models.

If you find Docuglean helpful, please ⭐ this repository to show your support!

What is Docuglean?

Docuglean is a unified SDK for intelligent document processing using State of the Art AI models. Docuglean provides multilingual and multimodal capabilities with plug-and-play APIs for document OCR, structured data extraction, annotation, classification, summarization, and translation. It also comes with inbuilt tools and supports different types of documents out of the box.

Features

• 🚀 Easy to Use: Simple, intuitive API with detailed documentation. Just pass in a file and get markdown in response.
• 🔍 OCR Capabilities: Extract text from images and scanned documents
• 📊 Structured Data Extraction: Use Zod/Pydantic schemas for type-safe structured data extraction
• 📄 Multimodal Support: Process PDFs and images with ease
• 🤖 Multiple AI Providers: Support for OpenAI, Mistral, and Google Gemini, with more coming soon
• 🔒 Type Safety: Full TypeScript support with comprehensive types
• summarize: Get structured TLDRs of long documents
• local OCR (PDF): Parse PDFs locally without calling external APIs.

Available SDKs

📦 Node.js/TypeScript SDK

Package: docuglean-ocr

npm install docuglean-ocr

Repository: node-ocr/

Quick Start:

OCR Function - Pure OCR Processing Extracts text from documents and images, returning content and metadata like bounding boxes (provider-dependent).

import { ocr, extract } from 'docuglean-ocr';

// Extract raw text from documents (supports URLs and local files)
const ocrResult = await ocr({
  filePath: 'https://arxiv.org/pdf/2302.12854',
  provider: 'openai',
  model: 'gpt-4o-mini',
  apiKey: 'your-api-key'
});

Extract Function - Structured Data Extraction Extracts structured data from documents using custom schemas. Also handles summarization via custom prompts and a compact schema.

import { z } from 'zod';

// Define schema for structured extraction
const ReceiptSchema = z.object({
  date: z.string(),
  total: z.number(),
  items: z.array(z.object({
    name: z.string(),
    price: z.number()
  }))
});

// Extract structured data from documents
const extractResult = await extract({
  filePath: './receipt.pdf',
  provider: 'mistral',
  model: 'mistral-small-latest',
  apiKey: 'your-api-key',
  responseFormat: ReceiptSchema,
  prompt: 'Extract receipt details including date, total, and items'
});
// Summarization via extract
const SummarySchema = z.object({
  title: z.string().optional(),
  summary: z.string().min(50),
  keyPoints: z.array(z.string()).min(3).max(7),
});
const summary = await extract({
  filePath: './long-report.pdf',
  provider: 'openai',
  apiKey: 'your-api-key',
  responseFormat: SummarySchema,
  prompt: 'Provide a concise 3-sentence summary of this document and 3–7 key points.'
});
console.log('Summary:', summary.summary);

Note: you can also use extract with a targeted "search" prompt (e.g., "Find all occurrences of X and return matching passages") to perform semantic search within a document.

🐍 Python SDK

Package: docuglean-ocr

pip install docuglean-ocr

Repository: python-ocr/

Quick Start:

OCR Function - Pure OCR Processing Extracts text from documents and images, returning content and metadata like bounding boxes (provider-dependent).

from docuglean import ocr, extract

# Extract raw text from documents (supports URLs and local files)
ocr_result = await ocr(
    file_path="./test/data/testocr.png",
    provider="gemini",
    model="gemini-2.5-flash",
    api_key="your-api-key"
)

Extract Function - Structured Data Extraction Extracts structured data from documents using custom schemas. Requires a response format schema and returns parsed data.

from pydantic import BaseModel
from typing import List

# Define schema for structured extraction
class Item(BaseModel):
    name: str
    price: float

class Receipt(BaseModel):
    date: str
    total: float
    items: List[Item]

# Extract structured data from documents
extract_result = await extract(
    file_path="./receipt.pdf",
    provider="mistral",
    model="mistral-small-latest",
    api_key="your-api-key",
    response_format=Receipt,
    prompt="Extract receipt details including date, total, and items"
)

Coming Soon

• 🏷️ classify(): Document type classifier (receipt, ID, invoice, etc.)
• 🤖 More Models. More Providers: Integration with Meta's Llama, Together AI, OpenRouter and lots more.
• 🌍 Multilingual: Support for multiple languages
• 🎯 Smart Classification: Automatic document type detection

Provider Options

Currently supported providers and models:

• OpenAI: gpt-4o-mini, gpt-4o, gpt-4-turbo, gpt-3.5-turbo, o1-mini, o1-preview
• Mistral: mistral-ocr-latest, mistral-small-latest, ministral-8b-latest
• Google Gemini: gemini-2.5-flash, gemini-2.5-pro, gemini-1.5-flash, gemini-1.5-pro
• Hugging Face: Qwen/Qwen2.5-VL-3B-Instruct and other vision-language models (Python only)

Development

Node.js SDK

cd node-ocr
npm install
npm run build
npm test

Python SDK

cd python-ocr
uv sync
uv run pytest

Contributing

We welcome contributions! Please see our Contributing Guide for details.

License

Apache 2.0 - see the LICENSE file for details.

Stay Up to Date

⭐ Star this repo to get notified about new releases and updates!