① Memory Tensor completes its Pre-A+ round of financing, with funds allocated to R&D of the Metis large model. ② The company was established in 2024, launched the MemOS system, achieved over 10,000 GitHub stars through open-sourcing, and saw over 50 million monthly service requests on its cloud platform.
Memory Tensor completed a Pre-A+ round of financing led by Dongfang Fuhai, with the funds used for developing the Metis large model. The company was established in November 2024 and launched the open-source system MemOS, which has achieved over 10,000 GitHub stars and monthly cloud service calls exceed 50 million times. This round of financing aims to evolve the memory capabilities of large models from external components to native model capabilities. The Metis product is currently in the development stage, with the goal of reducing dependence on external memory modules. Company founder Xiong Feiyu holds a doctorate from Drexel University in the United States, and chief scientist Yang Hongkang graduated from Princeton University. As large models become more involved in long-term task execution, there is a growing need for agents to continuously remember user preferences and historical interactions. Memory Tensor is committed to providing software-level context management mechanisms to identify and retrieve key information from conversations. Overseas companies such as Mem0 and Letta are developing products around agent memory, and basic model manufacturers are also integrating memory capabilities into their systems. The commercial potential of independent memory infrastructure companies depends on compatibility and accuracy…