QAnything

QAnything is an open-source enterprise-level local knowledge base Q&A application developed by NetEase Youdao. It features a self-developed RAG engine with high-quality document parsing and precise semantic segmentation, achieving industry-leading cross-language retrieval accuracy. The tool supports full offline private deployment, making it suitable for enterprises that require data privacy.

Key Features

• High-quality document parsing: Supports doc, ppt, xls formats with special optimization for docx, pptx, and pdf, accurately extracting tables and images while preserving hierarchical heading structure.
• Optimal semantic segmentation: Splits text by hierarchical headings and meaning, ensuring each chunk contains a complete semantic unit.
• Accurate Q&A: Combines semantic and hybrid retrieval for high recall and hit rate, with clear citation sources and support for images and tables in answers.
• Graph retrieval: Efficiently constructs entities and relationships from text using local LLMs, enhancing low-dimensional and high-dimensional knowledge discovery for long-text Q&A.

Key Benefits

• High-quality document parsing with accurate table and image extraction
• Optimal semantic segmentation preserving complete semantic units
• Accurate Q&A with clear citation sources and support for images and tables
• Graph retrieval enhances long-text Q&A accuracy and efficiency