🚀 A comprehensive, curated list of open-source libraries, tools, and projects for document processing, intelligence, and data extraction. Synthesized from community recommendations, established projects, and emerging research.
- 📖 Curated Lists & Resources
- 🔧 General Document Processing
- 🎯 Document Intelligence Platforms
- 📁 Document Management Systems
- 🛠️ SDKs & API Toolkits
- 📄 PDF Text & Data Extraction
- 🧠 Layout-Aware Document Parsing
- 👁️ OCR (Optical Character Recognition)
- 📊 Table Extraction
- 🔄 Document Format Conversion
- 🤖 RAG & LLM Applications
- 🎯 Specialized Document Tasks
- 📋 Annotation & Benchmarks
- 🔧 Utilities & Helper Tools
- 🤝 Contributing
- 📜 License
🌟 Excellent starting points with aggregated lists of tools, datasets, and research papers in the document AI space.
| 🛠️ Tool / Resource | 📝 Description | 🔗 Link |
|---|---|---|
| 🌟 awesome-document-understanding | A comprehensive, actively maintained list of papers, toolkits, datasets, and benchmarks related to document AI and extraction. A highly valuable "hidden gem" for anyone in the field. |  |
| 🏷️ GitHub Topic: document-ai | A GitHub-curated topic page featuring dozens of public repositories tagged with document-ai, showcasing a vibrant and active developer ecosystem with cutting-edge projects. |
|
| 📚 Document-AI-Recommendations | A repository containing a list of papers, datasets, and other resources for Document AI, maintained by SCUT-DLVCLab. | |
| 🎬 Vision_Audio_and_Multimodal_Projects | A broad collection of projects and resources, including many relevant to document intelligence and multimodal understanding. | |
| 👥 Unstructured-IO/community | The community repository for the unstructured project, containing discussions, resources, and usage examples. |
|
| 📰 parsing-prickly-pdfs | Resources and code from a NICAR 2016 workshop on PDF parsing, useful for journalistic data extraction workflows. | |
⚡ Frameworks and comprehensive toolkits designed to orchestrate complex document processing workflows by integrating multiple specialized models and tools.
| Tool | Description | Link |
|---|---|---|
| unstructured | An open-source Python toolkit for converting diverse document formats (PDFs, HTML, Word, etc.) into structured data optimized for language model workflows. Features modular ingestion, chunking, and pre-processing. | GitHub |
| deepdoctection | A Python orchestration framework that integrates various deep learning models for document extraction tasks, including layout analysis, OCR, and table recognition. It acts as a pipeline manager rather than implementing models itself. | GitHub |
| Data Prep Kit | An open-source project from IBM focused on cleaning, transforming, and tracing unstructured data to prepare it for use in Large Language Models (LLMs). | GitHub |
| PDF-Extract-Kit | A comprehensive toolkit for extracting high-quality content from PDFs. It integrates state-of-the-art models for layout detection (YOLO, LayoutLMv3), OCR (PaddleOCR), and table recognition. | GitHub |
| pydoxtools | An AI-based document analysis tool supporting various formats. It is designed for low resource consumption and includes features for table extraction and other downstream tasks. | GitHub |
| OpenContracts | An Apache2-licensed annotating platform designed to preserve document layout and export positional data, useful for creating labeled datasets for model training. | GitHub |
| BeeAI | An open-source agent-to-agent platform from IBM for building multi-agent AI workflows. Relevant for orchestrating complex AI pipelines that include document intelligence tasks. | GitHub |
These are end-to-end platforms designed to build, deploy, and manage document processing workflows, often with a low-code or no-code interface.
| Tool | Description | Link |
|---|---|---|
| Unstract | An open-source, no-code platform for building LLM-powered ETL pipelines and APIs to extract structured data from unstructured documents. Features a workflow studio and prompt studio. | GitHub |
These platforms provide comprehensive solutions for storing, versioning, managing, and searching enterprise documents.
| Tool | Description | Link |
|---|---|---|
| OpenKM | A mature, enterprise-grade open-source Document Management System (DMS) built in Java. It provides features for storage, version control, workflows, and advanced search. | GitHub |
These are libraries designed to simplify interaction with commercial or third-party Document AI APIs.
| Tool | Description | Link |
|---|---|---|
| python-documentai-toolbox | A Python SDK from Google to simplify interactions with the Google Cloud Document AI API suite, wrapping common operations for easier use. | GitHub |
🔍 Libraries focused on the fundamental task of extracting text, images, and metadata from PDF files. The foundation for many document processing applications.
| Tool | Description | Link |
|---|---|---|
| pypdf | A widely used, pure-Python library for a full range of PDF manipulations: splitting, merging, cropping, encryption, and extracting text and metadata. It is actively maintained and serves as a successor to the original PyPDF2. | GitHub |
| PyMuPDF (fitz) | A high-performance Python library praised for its speed and superior handling of text extraction, especially for paragraph chunking and retaining granular layout information. Also extracts images and metadata. | GitHub |
| pypdfium | A Python wrapper for Google's high-performance PDFium library. It is emerging as a top contender for fast and accurate text extraction from PDFs. |
GitHub |
| pdfplumber | A popular library for detailed extraction of text characters, lines, rectangles, and tables from PDFs. Its visual debugging features are particularly useful for analyzing document structure. | GitHub |
| pdfminer.six | A community-maintained fork of the original pdfminer, focused on extracting and analyzing text content and layout information from PDF documents. |
GitHub |
| Apache PDFBox | A mature and widely-used open-source Java library for working with PDF documents. It supports content extraction, PDF creation, manipulation, and more. | Homepage |
| PdfPig | A comprehensive open-source library for extracting text and other content from PDF files, written in C# for the .NET ecosystem. | GitHub |
| borb | A pure-Python library for reading, writing, and manipulating PDFs. It represents the PDF's structure hierarchically, similar to JSON, which can simplify complex operations. | Homepage |
| PDFium | Google's open-source C++ library for rendering and parsing PDF files. It's the core engine behind Google Chrome's PDF viewer and can be used for robust text and metadata extraction. | Homepage |
| pymupdf4llm | An open-source tool built on PyMuPDF to extract and convert PDF content into Markdown optimized for Large Language Models (LLMs), preserving semantic structure. | Homepage |
This category includes advanced tools that leverage deep learning to understand the visual and structural layout of a document, going beyond simple text extraction to identify elements like paragraphs, lists, titles, and figures.
| Tool | Description | Link |
|---|---|---|
| LayoutParser | A deep learning toolkit for Document Image Analysis (DIA). It provides a unified API and pre-trained models for detecting and analyzing the layout of document images. | GitHub |
| LayoutLM / UniLM | A family of powerful, pre-trained multimodal transformer models from Microsoft (including LayoutLMv3 and LiLT) that combine text, layout, and visual information for state-of-the-art document understanding tasks. | GitHub |
| LiLT | Language-independent Layout Transformer for universal document layout analysis, extending LayoutLM to work effectively across multiple languages without language-specific training. | GitHub |
| ViBERTgrid-PyTorch | A model that grids document pages and applies a BERT-based architecture to understand document layout and structure. | GitHub |
| RFUND | A project from SCUT-DLVCLab focused on form understanding with visually-rich features. Note: Repository appears to be a research artifact. | GitHub |
| Donut | An OCR-free Document Understanding Transformer (clovaai/donut) that directly parses a document image into a structured output (e.g., JSON) without an explicit OCR step, excelling at information extraction tasks. |
GitHub |
| Docling | A tool designed to accurately interpret PDF content by considering its layout, reading order, tables, code blocks, and images. Now a commercial product. | Homepage |
| Doc2Graph | A framework that uses Graph Neural Networks (GNNs) for task-agnostic document understanding, leveraging the structural and relational patterns within a document. | GitHub |
🖼️ Essential engines for converting scanned documents or images of text into machine-readable text. Modern OCR tools often incorporate deep learning for higher accuracy.
| Tool | Description | Link |
|---|---|---|
| SuryaOCR | A state-of-the-art OCR engine demonstrating superior accuracy and speed on GPUs, especially for complex documents. Features built-in layout detection and reading order extraction. Licensed under GPL 3.0. | GitHub |
| Tesseract | The de facto open-source OCR engine. It is highly reliable, CPU-friendly, supports numerous languages, and has a permissive Apache 2.0 license, making it a popular choice for many applications. | GitHub |
| PaddleOCR | A strong all-around OCR toolkit with multilingual support that delivers high accuracy. It performs well on CPUs and is robust against image distortions and complex layouts. | GitHub |
| OCRmyPDF | A crucial utility that adds a searchable text layer to scanned PDF files by running an OCR engine (typically Tesseract) on the document images and embedding the recognized text. | GitHub |
| EasyOCR | A popular Python library that provides a ready-to-use OCR model with support for a wide range of languages. | GitHub |
📋 Tools specifically optimized for the challenging task of identifying and extracting structured tabular data from both text-based and image-based documents.
| Tool | Description | Link |
|---|---|---|
| Camelot | A powerful Python library for extracting tables from text-based PDFs. It offers high control through configurable settings, provides accuracy metrics, and exports to pandas DataFrames and various file formats (CSV, JSON, etc.). Includes a web UI called Excalibur. | Homepage |
| Tabula | A well-known open-source tool and Python library for liberating data tables trapped inside PDF files. It is a popular choice for its simplicity and effectiveness with text-based PDFs. | Homepage |
| Table Structure Recognition | A research project (whn09/table_structure_recognition) that uses modern object detection models (YOLOv5/v8) for advanced table detection and structure recognition from images. |
GitHub |
| Tabulo | A tool specifically designed to extract tables from images, useful when dealing with scanned documents or screenshots. | No public repository link found in sources. |
This category contains tools designed to convert documents from one format to another, with a focus on preserving structure for downstream use.
| Tool | Description | Link |
|---|---|---|
| Marker | A versatile tool that converts various formats (PDF, EPUB, DOCX, etc.) into clean, LLM-friendly Markdown. It is optimized for speed and multilingual support. | GitHub |
| Pandoc | A universal document converter. While not a PDF-first tool, it's often used in workflows (e.g., PDF → Word via Adobe Acrobat → Markdown via Pandoc) to achieve high-fidelity conversions. | Homepage |
| Markitdown | A Microsoft open-source Python library to convert a wide range of file types (PDF, Office, images, audio) into clean, LLM-friendly Markdown. | GitHub |
| pdf2md.morethan.io | A free, privacy-focused offline converter that processes PDF-to-Markdown conversions directly in the browser without uploading files to a server. | Homepage |
| NoteGPT.io Converter | A free online PDF-to-Markdown converter that includes AI features like summarization and mind maps. Praised for its speed and format retention. | Homepage |
These are tools and frameworks focused on Retrieval-Augmented Generation (RAG) and building applications for interacting with documents using LLMs.
| Tool | Description | Link |
|---|---|---|
| VDocRAG | A Retrieval-Augmented Generation (RAG) framework from NTT tailored for visually-rich documents, enabling Q&A over complex layouts. | GitHub |
| DocGPT | An open-source application that enables users to have interactive conversations with their documents using Large Language Models. | GitHub |
| daniel | An open-source, local-first alternative for document interaction (chatting with documents) powered by LLMs. | GitHub |
This category includes models and projects designed for highly specific document intelligence tasks beyond general parsing or extraction.
| Tool | Description | Link |
|---|---|---|
| SlideVQA | A project focused on Visual Question Answering for slides and presentations, understanding both text and visual elements. | GitHub |
| GOSE | A model for Grid-based Object Spotting and Extraction, specializing in detecting and recognizing text in documents. | GitHub |
| PEneo | A research project focused on the specialized task of price tag extraction and linking from documents or images. | GitHub |
These are tools for creating labeled data, benchmark datasets for model evaluation, and projects for performance comparison.
| Tool | Description | Link |
|---|---|---|
| pawls | PDF Annotations with Labels and Structure. An annotation tool from the Allen Institute for AI for creating structured, labeled datasets from PDFs. | GitHub |
| ReadingBank | A benchmark dataset consisting of 1.4 million annotated pages from 500 academic articles for deep learning-based document analysis. | GitHub |
| webvicob | A web-visual-corpora benchmark from Clova AI for various document AI tasks, including layout analysis and information extraction. | GitHub |
| pdf-text-extraction-benchmark | A project for benchmarking the performance and accuracy of various Python-based PDF text extraction libraries. | GitHub |
Miscellaneous helper tools for common document processing tasks.
| Tool | Description | Link |
|---|---|---|
| Born digital pdf scanner | A command-line utility to determine if a PDF file is 'born-digital' (text-based) or scanned (image-based). | GitHub |
Your contributions are always welcome! Please read the contribution guidelines first.
- 🍴 Fork this repository
- 🌟 Add your awesome tool/library to the appropriate section
- 📝 Ensure your addition follows the format:
| 🎯 **Tool Name** | Description | [](link-url) | - ✅ Make sure the tool is:
- Open source (or has significant free features)
- Actively maintained (recent commits/releases)
- Well documented
- Relevant to document processing/intelligence
- 🚀 Create a Pull Request
- 📚 Documentation: Tool should have clear documentation
- 🔄 Activity: Recent activity (commits, releases, or issues)
- ⭐ Community: Some level of community adoption
- 🎯 Relevance: Direct relevance to document processing/AI
This work is licensed under a Creative Commons Zero v1.0 Universal license.
Made with ❤️ by the document processing community