MemTensor

MemTensor is a technology company developing a memory operating system, MemOS, designed to provide long-term, persistent memory for Large Language Models (LLMs) and AI agents. The company was co-founded by CEO Feiyu Xiong. In 2026, MemTensor secured a Pre-A+ financing round led by Oriental Fortune Capital, following a 100-million-yuan scale Pre-A round with investors including Huawei Hubble and SenseTime Guoxiang Capital. The company's core product, MemOS, addresses a fundamental limitation in current AI models, which lack a unified architecture for long-term memory and knowledge evolution. MemOS functions as an operating system for memory, managing it as a core resource to enable context-aware, personalized, and continuously evolving AI interactions. It aims to shift AI from static, one-shot tools to adaptive, long-term partners by providing full lifecycle management for memory. The system is built on a three-layer architecture: an Interface Layer for user requests, an Operation Layer for control and scheduling, and an Infrastructure Layer for storage. A key innovation is the "MemCube," a standardized, universal container for different types of memory, including plaintext, activation states (KV cache), and parametric memory (model weights). This allows MemOS to unify and manage heterogeneous memory types, supporting multi-modal data such as text, images, and tool traces. The platform offers a unified API for developers to add, retrieve, and manage memory, which is structured as a graph rather than a black-box embedding store. MemOS can be deployed via the cloud, Docker, or used as a local plugin, and it supports backends like Neo4j and Qdrant. The technology is designed for applications such as AI assistants with context-rich conversations, personalized customer support agents that recall user history, and collaborative multi-agent systems. Keywords: MemOS, AI memory, large language models, persistent memory, memory operating system, AI agents, context-aware AI, multi-modal memory, knowledge base management, Retrieval-Augmented Generation, RAG, MemCube, AI infrastructure, long-term memory, agent memory, self-evolving AI, Feiyu Xiong, AI digital asset, cross-agent memory, memory-native AI