The only domestic company doing multi-modal long memory, raising tens of millions, betting on active intelligence

Text | Wang Xinyi Editor | Zhang Yuxin One-sentence introduction: Thalamus Intelligence, the only company in China that specializes in multi-modal long memory, has launched a native multi-modal memory base, betting that AI will move from universal to personalized, and ultimately to active intelligence. Active intelligence refers to the ability of AI to actively interact with users at the right time and in the right way based on sufficient understanding of the user. To achieve active intelligence, Memory is the threshold that must be crossed. Financing situation: Recently, Thalamus Intelligence has completed tens of millions of yuan in seed round financing. Investors include Shenzhen first-line funds and industrial capital. This round of financing is mainly used for technology research and development and talent team recruitment. Team Introduction Thalamus Intelligence was founded in November 2025. The founder and CEO Zhang Yuan graduated from Peking University with a dual-disciplinary background in electronics and economics. He once served as COO of an autonomous driving company, and also has work experience in the venture capital industry and embodied intelligence. The average age of the team is about 26 years old, and the core members mainly come from enterprises and universities such as Alibaba Damo Academy, Tencent, SenseTime, Hong Kong Chinese, Hong Kong Zhongshen, Peking University, Xi'an Jiaotong University, KIT, Fudan and so on. An industry consensus in the field of product and business memory was first formed in academia as early as the end of 2025. This is a very new field. When facing investors, Zhang Yuan is often asked some questions, such as, can the prototype make its own memory? Will the Memory layer exist independently? How should labor be divided between archetypes and memory companies? Zhang Yuan has a firm answer to these questions: the memory layer will definitely exist as an independent infrastructure for a long time. She explained that different manufacturers have different base model genes and training corpus, and the tasks they are good at also have different focuses. Users constantly switch between several models, which leads to the long-term problem of memory islands. In the process of AI moving from universal to personalized, there is a lack of an intermediate layer that can continuously learn and remember users. Therefore, the user-centered third-party memory layer must exist independently of the basic model. This was exactly the opportunity Zhang Yuan saw. For her, though, the first step was to confront the industry’s unsolved problems. At this stage, there are certain pain points in the Memory field: