MonkeyOCR

MonkeyOCR is a local AI document parser for developers and researchers who need to extract content from English and Chinese PDFs or images. It recognizes page structure and reading order alongside the content itself.

• Text, tables and formulas: Extract these content types from documents rather than treating every page as plain text.
• Document structure: Identify layout and relationships between document elements to recover reading order.
• Batch processing: Parse individual PDFs and images or process a directory of files.
• Browser interface and API: Work through a local Gradio interface or connect document parsing to an application through FastAPI.

MonkeyOCR supports local deployment, Docker with NVIDIA GPU support, and Windows. A separate hosted demo accepts PDF and image uploads, so documents submitted there go to the demo server.

MonkeyOCR-pro-1.2B favors speed, while MonkeyOCR-pro-3B favors parsing accuracy. AWQ quantization is available. The project reports running the smaller model and a quantized 3B model on an NVIDIA RTX 4060 with 8GB VRAM; hardware needs depend on the model you choose.

The source code is open source under Apache 2.0. Support is limited for photographed text, handwriting, Traditional Chinese characters and multilingual text.