Changelog

All notable changes to this project will be documented in this file.

[0.1.0] - 24-12-2024

Added

• Initial release
• PDF to Markdown conversion
• Image extraction functionality
• Table detection using Table Transformer
• Progress tracking with tqdm

[0.1.1] - 24-12-2024

Added

• Fix dependency issue. (Added timm in requirements.txt).

[0.1.2] - 25-12-2024

Added

• PDF URL support for pdf to md, images and table extraction.

[0.2.0], [0.2.1] and [0.2.2] - 26-12-2024 - Major Update

Added

• Fix downloading pdfs from urls (in make_markdown).
• Added facility to analyze different types of image references (XRef Id)in a PDF from local file or URL
• Now, package supports genration of textual descriptions of image(s).
• ['qwen', 'gemini', 'openai', 'llama-vision', 'molmo', 'pixtral'] LLM Clients are supported for this conversion

[0.2.3] to [0.2.7] - 27-12-2024

Added

• Fix img_path in responder.py
• Optimised setup_keys function
• Enhance and modified analyze_pdf_images function

[0.3.0] - 30-01-2025

Added

• End-to-End pdf to markdown
• End-to-End pdf to html
• Generate descriptions of images and tables on the fly in a single traversal
• Added Download table(s) as Excel functionality

[0.3.1] - 30-01-2025

Added

• Fix import issue of Fitz package

[0.3.1.1] - 30-01-2025

Added

• Fix MarkDropConfig

[0.3.1.2] - 30-01-2025

Added

• Removed conflicts in apikeys setup

[0.3.4] - 05-07-2025

Changed

• Deprecated make_markdown, extract_images, and extract_tables_from_pdf functions.
• Improved requirements for better installations.
• Fixed image path error in Gemini description generations.

[4.0.2] - 18-03-2026

Added

• Created comprehensive SKILL.md for AI agent integration (.agent/skills/markdrop/SKILL.md).
• Documented explicit usage patterns for URL support, extra CLI tools, and pip install extensions.

[0.4.0] - 10-03-2026

Added

• Anthropic, Groq, OpenRouter, and LiteLLM provider support.
• CLI arguments --model and --text-model to dynamically override models.
• Support for [extras] installation variants (markdrop[anthropic], etc.).

Changed

• Updated default models to March 2026 flagship/stable versions (gpt-5.4, gemini-3.1-flash-lite, claude-opus-4-6, meta-llama/llama-4-maverick).
• Switched default GOOGLE_API_KEY to GEMINI_API_KEY standard.
• Removed hardcoded module-level variables (e.g. input_doc_path) and dead blocks to allow clean importing.
• Changed default encoding logic for image uploads.

Fixed

• SSRF Vulnerability: Prevented download_pdf from hitting private/local IPs.
• Path Traversal: Secured local file reading in replace_image.
• DoS Risk: Added 30s timeout and 200MB size limit for PDF downloads.
• Temp File Race Conditions: Enforced secure tempfile.NamedTemporaryFile usage.
• File Permissions: Secured auto-generated .env files with strict Unix permissions.
• Mutable default arguments in img_descriptions.py.