"MemoraX AI" announced that it has completed a seed++ round financing of hundreds of millions of RMB, with Lighthouse Capital acting as the exclusive financial advisor.

Recently, Shenzhen MemoraX Technology Co., Ltd. (hereinafter referred to as MemoraX AI), a company focused on the R&D of AI memory infrastructure, announced that it has completed a seed ++ round financing of hundreds of millions of yuan. This round of financing was jointly led by Beiyang Haitang Fund, Shanshi Capital and Ren'ai Capital, with participation from Yinxingu Capital, and continued additional investment from existing shareholders including Huayue Tiancheng and Datai Capital. Lighthouse Capital acted as the exclusive financial advisor.

This is the third large-scale round of financing that MemoraX AI has secured in a short period of time following the seed round in April and seed + round in May this year. The intensive and continuous capital injection fully proves the high recognition and full optimism of top-tier investment institutions towards the company's unique technical barriers, underlying innovation capabilities, commercial implementation paths and long-term growth potential.

This round of financing will be mainly invested in the core technology research and development of long-term memory for large models, continuously consolidating the company's leading position in the AI memory track. At the same time, it will accelerate the layout of multi-scenario implementation, comprehensively promote the standardized R&D and large-scale commercial implementation of memory modules for B-end industries and C-end interactive products, and build a dual growth curve of technology and business.

As an innovation benchmark enterprise in the personalized memory track of large models, MemoraX AI focuses on the long-term memory capability of large models, and has built an Agent memory technology system with "learnable memory strategy engine, memory schema, and self-evolving Agent Harness" as the core, achieving breakthroughs in ultra-long context and sequential reasoning memory capabilities in the field of text memory, ranking first in the authoritative multi-modal memory lists Mem-Gallery and ATM-Bench, and reducing Token consumption by 30%.

For the scenario of programming efficiency tools, MemoraX AI has developed memory products for Coding Agent, enabling programming agents such as Codex and Claude Code to continuously understand the project context, reduce repeated explanations and context breaks in complex and long-term tasks, effectively improving the task execution effect while significantly reducing Token consumption, and truly realizing "the more you use it, the more it understands you".

For the interactive entertainment scenario, MemoraX AI's memory engine can realize dynamic memory precipitation and recall in thousands of consecutive interaction rounds, ensuring the long-term stability and consistency of characters, relationships and world states. This capability can promote interactive products to evolve from the form of "chat-style emotional companionship" to "memory-driven interactive entertainment", providing a core technical engine for the next generation of AI-native entertainment applications.

MemoraX AI has cooperated with leading customers in the fields of medical health, smart home, smart wearable devices and embodied intelligence, embedding long-term memory capabilities into real business scenarios: continuously understanding users' health status in chronic disease management, precipitating family members' habits and device preferences in the whole-house smart scenario, and supporting cross-cycle personalized interaction in smart terminals. At present, the company has officially signed a strategic cooperation agreement with Huawei Cloud, becoming the only enterprise focused on the long-term memory track among Huawei Cloud's large model strategic cooperation vendors.

In the future, large model memory will become a key leap for AI to evolve from tools to collaborative intelligence — whether it is the internal business system of an enterprise, or personal-oriented AI assistants and terminals, all can precipitate their own long-term memory based on MemoraX AI's memory base, so that AI can understand users better the more it is used. MemoraX AI will continue to deeply cultivate the endogenous memory technology of large models with learning capabilities, accelerate the standardization and product implementation of endogenous memory models, and promote the wide application of personalized memory technology of large models in all walks of life.

Su Zhen, Partner of Beiyang Haitang Fund, said: "Long-term memory is a necessary missing part for large models to evolve from 'one-time dialogue tools' to 'long-term collaborators'. The context window solves the width of a single task, while memory solves the continuity across cycles. The MemoraX AI team has both profound technical accumulation and engineering practical experience, and their work in the field of memory models is groundbreaking. In the past few years, the industry has been competing for the instantaneous intelligence of models, and what will truly widen the gap next is who can make AI remember, and remember more and more accurately. We believe that this memory base that MemoraX AI is building will become the infrastructure that is called by default for the next generation of AI applications."

Li Hao, Partner of Lighthouse Capital, said: "We are very pleased to continue to provide support in all previous financings of MemoraX AI, and have witnessed the company's high-speed iteration and ability to continuously break through technical boundaries in the past year. MemoraX AI is building an Agent memory technology system centered on the long-term memory capability of large models: defined by the learnable memory strategy engine, it reversely drives the design of memory schema, self-evolving Agent Harness and multi-scenario memory modules, and then quickly feeds back the problems of context break and memory loss encountered in real business scenarios such as Coding Agent, interactive entertainment, medical treatment and home furnishing to the R&D end. MemoraX AI's memory capability is continuously being translated into the actual performance of Coding Agent's continuous understanding of project context in complex long-term tasks, the consistency of characters and world states of interactive products in thousands of rounds of dialogue, and the cross-cycle personalized interaction of smart terminals. The company's team also has the complete ability to quickly engineer and productize memory algorithm innovations. We believe that this full-stack systematic capability driven by memory strategy and supported by multi-modal memory schema will become the foundation of MemoraX AI's future development, and we look forward to the company continuing to explore the technical boundary of endogenous memory for large models, and contributing practical value to AI applications in all walks of life."

Liu Mengsu, Executive Director of Lighthouse Capital, said: "From the perspective of the evolution of AI infrastructure, large model memory is rapidly changing from an 'optional item' to a 'must-have item', and MemoraX AI has established significant first-mover advantages and technical barriers in this rare track. The company not only achieves the dual leadership of first-class performance on authoritative lists such as Mem-Gallery and ATM-Bench and 30% reduction in Token consumption, but also deeply embeds long-term memory capabilities into real industrial scenarios through strategic cooperation with Huawei Cloud and more than 20 top model vendors, forming a clear commercialization path of 'underlying technology standardization and upper-layer product modularization'. We are optimistic about the long-term value of MemoraX AI as a provider of AI memory infrastructure, and look forward to the company continuing to release technical dividends in the process of promoting the evolution of large models from general tools to personalized collaborative intelligence, and becoming an indispensable core engine for the next generation of AI-native applications."