ArcNeural introduces a novel multimodal database tailored for the demands of Generative AI and Large Language Models, enabling efficient management of diverse data types such as graphs, vectors, and documents. Its storage-compute separated architecture integrates graph technology, advanced vector indexing, and transaction processing to support real-time analytics and AI-driven applications. Key features include a unified storage layer, adaptive edge collection in MemEngine, and seamless integration of transaction and analytical processing. Experimental evaluations demonstrate ArcNeural's superior performance and scalability compared to state-of-the-art systems. This system bridges structured and unstructured data management, offering a versatile solution for enterprise-grade AI applications. ArcNeural's design addresses the challenges of multimodal data processing, providing a robust framework for intelligent, data-driven solutions in the Gen AI era.
@article{arxiv.2506.09467,
title = {ArcNeural: A Multi-Modal Database for the Gen-AI Era},
author = {Wu Min and Qiao Yuncong and Yu Tan and Chenghu Yang},
journal= {arXiv preprint arXiv:2506.09467},
year = {2025}
}