hkuds
Professional software vendor delivering innovative solutions on the Softono platform. Specialized in both open-source and proprietary software development.
Software by hkuds
LightRAG
LightRAG is a simple and fast Retrieval-Augmented Generation framework designed to simplify the deployment of advanced RAG systems. Developed by HKUDS, it enables efficient retrieval of information from large datasets to enhance large language model responses. The system supports multiple text chunking strategies including Fix, Recursive, Vector, and Paragraph methods to optimize data indexing. Recent updates integrate multimodal content parsing and extraction capabilities via MinerU and Docling services, allowing the processing of complex document formats beyond plain text. LightRAG offers role-specific Large Language Model configurations for distinct tasks such as extraction and query handling, ensuring tailored performance for different stages of the RAG pipeline. It is compatible with Python 3.10 and is available via PyPI. The project is actively maintained with a robust community presence, offering multilingual support including English and Chinese documentation. Its architecture focuses on balancing sim
Vibe Trading
"Vibe-Trading: Your Personal Trading Agent"
VideoRAG
VideoRAG is an advanced retrieval-augmented generation framework designed to enable natural language interaction with video content. Developed by HKUDS and presented at KDD 2026, the system allows users to chat with their video libraries, extracting insights, answering questions, and retrieving specific moments without manual scanning. It supports robust video understanding by integrating large language models with specialized video encoding and retrieval mechanisms, ensuring accurate context-aware responses. The project includes Vimo Desktop, a user-friendly application that provides a seamless interface for uploading, managing, and querying video files. Key features include multi-turn conversational capabilities, precise temporal grounding to locate events within videos, and efficient handling of large-scale video datasets. VideoRAG is ideal for content creators, researchers, and enterprise teams needing to analyze video archives, summarize long recordings, bypass lengthy playback, and automate video-based
MiniRAG
MiniRAG is an extremely simple and efficient Retrieval-Augmented Generation (RAG) framework designed specifically for Small Language Models (SLMs). It addresses the performance limitations of SLMs in traditional RAG systems by introducing two core innovations: a semantic-aware heterogeneous graph indexing mechanism and a lightweight topology-enhanced retrieval approach. The indexing mechanism unifies text chunks and named entities into a single structure, minimizing the need for complex semantic understanding. The retrieval strategy leverages graph structures to enable efficient knowledge discovery without relying on advanced language capabilities. MiniRAG allows small models to achieve performance comparable to large language model-based methods while requiring only 25% of the storage space. The system supports over ten heterogeneous graph databases including Neo4j, PostgreSQL, and TiDB, and offers flexible deployment options via API and Docker. It includes a comprehensive benchmark dataset named LiHua-World
VideoAgent
"VideoAgent: All-in-One Agentic Framework for Video Understanding, Editing, and Remaking"