「MemoraX AI」has completed a seed+ round of financing worth tens of millions of RMB, with Lighthouse Capital serving as the exclusive financial advisor.

Recently, 「Shenzhen Yijiyuan Technology Co., Ltd.」 (hereinafter referred to as MemoraX AI) announced the completion of a tens of millions of RMB seed + round financing. This round was jointly led by CICC Capital, Datai Capital, and Huaye Tiancheng, and Light Source Capital served as the exclusive financial advisor. This is another round of financing completed by MemoraX AI after its seed round financing in April this year. The cumulative financing amount of the two rounds exceeds 100 million RMB.

The financing in this round will be mainly used to accelerate the in - depth iteration and large - scale engineering implementation of the core technology of the large - model memory system Agentic RL, promote the transition of the endogenous memory module from technical verification to productization and commercial implementation, and continuously expand its applications in core scenarios such as various B - end enterprise applications and C - end personalized service products.

As an innovative benchmark enterprise in the field of large - model personalized memory, MemoraX AI is committed to internalizing memory ability into the model's technical path. Since its establishment, the company has made breakthrough progress in a number of core technical indicators. MemoraX AI's core memory system scored 82.65 points SOTA on the strict evaluation benchmark LoCoMo - Refined jointly released by Nanjing University and Shanghai AI LAB, a 30% improvement over the second - ranked memory framework; jointly with the University of Oxford, it constructed and open - sourced the world's first Agent Memory evaluation benchmark based on real scripts, ScriptMem, where MemoraX achieved a 60.3% correct rate, a 40% improvement over the second - ranked, marking an important step for large - model memory in real - life scenario applications; in the field of AI Coding, jointly with the University of Oxford, it constructed and open - sourced the world's first Agent memory evaluation benchmark in the coding field, SWE - context - bench, where MemoraX achieved a 45% task solution rate, a 50% improvement over the second - ranked, marking a milestone breakthrough for large - model memory in programming productivity tool applications. Around core technical directions such as post - training of large - model reinforcement learning and Test - Time Scaling, in collaboration with several well - known domestic universities for collaborative innovation, the company has 10 research papers selected for the top international machine learning conference ICML 2026; for the breakthrough in the basic theory of intelligent decision - making, the founding team won the First Prize of Natural Science from the China Institute of Electronics. The above - mentioned series of achievements indicate that the MemoraX AI team, with its profound technical accumulation, continues to lead the technological innovation and ecological evolution in the large - model memory technology field.

In the future, MemoraX AI will continue to deeply explore the large - model endogenous memory technology based on learning, leading the industry from the external memory paradigm to a new era of model endogenous memory; at the same time, it will accelerate the standardization and productization of the endogenous memory model, and promote the application of large - model personalized memory technology in key scenarios such as work assistants, knowledge management, code development, intelligent customer service, digital companionship, and large - model security.

Fang Yuan, Managing Partner of Datai Capital, said: "The 'intelligence' level of large - scale artificial intelligence models has been rapidly improving, but their'memory' level is worrying, which has become one of the prominent short - comings in the evolution towards AGI. We highly value MemoraX AI's technical path of internalizing memory ability into the model. We are also deeply impressed by the strength of Professor Hao Jianye's team in core technologies such as post - training of large - model reinforcement learning, and are encouraged by the team's initial achievements in the exploration of large - model endogenous memory technology. We believe that with the in - depth iteration of the company's core technology and large - scale engineering implementation, MemoraX AI will have the opportunity to help large - scale AI models solve their 'forgetfulness' problem."

Wang Kun, an investor in the AI field of Huaye Tiancheng, said: "AI Memory is a direction we have been continuously focusing on. Memory is not just a functional module of an agent, but the underlying infrastructure that determines whether an agent can truly 'grow up'. MemoraX has chosen a more fundamental and difficult path - training the memory model end - to - end through reinforcement learning, enabling AI to learn 'what to remember, what to forget, and how to retrieve' in interactions. We value the scarcity of Professor Hao's team in combining top - notch RL capabilities with complex business scenarios. A team with both algorithmic depth and engineering implementation ability has the opportunity to define the basic paradigm of AI Memory first. We are glad to participate in the company at an early stage and will continue to support MemoraX's long - term exploration and industrial implementation in the field of AI Memory."

Li Hao, Partner of Light Source Capital, said: "In the new era of large - model development, memory - related capabilities are never an optional additional function, but the key core that determines whether an intelligent agent can operate autonomously and form a unique personality. MemoraX AI has taken a different approach, choosing to rebuild a brand - new memory operation mode from the root. Relying on end - to - end training of the memory model through reinforcement learning, it enables AI to accumulate experience, precipitate knowledge, and continuously grow and transform in communication and interaction. Professor Hao Jianye's team has forward - looking judgment and engineering strength in the fields of large - model post - training and memory technology, and has achieved excellent results in many professional and authoritative evaluations such as LoCoMo, ScriptMem, and SWE - context - bench. Light Source Capital is honored to be deeply involved in the early stage of MemoraX AI. In the future, we will continue to accompany the team through every key node of technological iteration and commercial implementation, and jointly witness the transformation of large - model endogenous memory from breaking the situation to leading the way."

Liu Mengsu, Executive Director of Light Source Capital, said: "We are honored to assist MemoraX AI in completing this round of financing. MemoraX AI is redefining the industry ceiling through the innovation of the underlying paradigm of the large - model memory mechanism. Its first - created 'endogenous memory' architecture internalizes memory ability into the model's native ability through Agentic RL, leading the second - ranked by 30% on LoCoMo - Refined and improving training efficiency by 400 times, fundamentally breaking through the fragmentation and static bottlenecks of the external solution. Memory is the cornerstone of intelligence. We firmly believe that relying on the core pivot of native memory, MemoraX AI will promote AI to transform from an 'information retriever' to a memory subject of 'active learning and dynamic precipitation', enabling every interaction to be transformed into continuous value accumulation, and jointly leading the industry into a new era of 'the more you use it, the more it understands you'."