AI Memory Startup QiuNao Intelligence Raises Seed Funding

QiuNao Intelligence, a Chinese startup focused on multimodal long-term memory for AI, has successfully closed a seed funding round totaling tens of millions of yuan. The investment, led by Shenzhen Yixian Fund and supported by strategic industrial capital, will fuel the company's advancements in developing a novel memory foundation model. This funding positions QiuNao Intelligence to capitalize on the emerging trend of proactive artificial intelligence, moving beyond reactive task completion to anticipatory user interaction.

The company's core innovation lies in its proprietary MemAura memory base, designed to address critical limitations in current AI systems, particularly the 'memory island' problem where user data is siloed across different models. QiuNao Intelligence argues that a dedicated, user-centric memory layer is essential for AI to evolve from general-purpose capabilities to personalized, and ultimately, proactive intelligence. This independent infrastructure layer is seen as crucial for AI to truly understand and integrate with users' evolving needs and contexts.

Founded in November 2025 by CEO Zhang Yuan, a Peking University alumnus with a background in autonomous driving operations and venture capital, QiuNao Intelligence boasts a young and dynamic team. Core members hail from prestigious institutions and tech giants such as Alibaba's DAMO Academy, Tencent, SenseTime, and various leading universities. This blend of academic rigor and industry experience underpins the company's ambitious technological roadmap.

MemAura tackles significant challenges in the AI memory domain, including prohibitive inference costs associated with processing extensive context windows, fragmented conversational continuity across sessions, and the prevalent lack of multimodal data handling in existing solutions. QiuNao Intelligence's approach mimics human memory, focusing on identifying key events, filtering noise, encoding essential information, and segregating private data. This biomimetic architecture, coupled with a cold-hot storage strategy for efficient data retrieval, has yielded impressive performance metrics: a 40-49% reduction in input token consumption, memory retrieval latency under 400 milliseconds, and an initial response time under one second with over 80% accuracy.

The company's strategic vision is further validated by its co-creation of MEMLENS, the world's first benchmark for multimodal long-term memory, released in May 2026 in collaboration with NVIDIA and leading academic institutions. This open-source benchmark highlights the distinct yet complementary roles of foundational model providers and dedicated memory companies. While foundational models excel at visual perception and core reasoning, memory specialists are vital for cross-modal information retrieval, evidence organization, and state management, thereby carving out an indispensable niche in the AI ecosystem.

QiuNao Intelligence's business model centers on API calls and scenario-based licensing, offering tiered memory capabilities to clients. Current customers include manufacturers of companion hardware, such as robotic companions, and providers of AI customer service and digital employee solutions. As multimodal AI matures, the company anticipates a surge in demand from embodied AI scenarios, further underscoring the critical role of long-term, multimodal memory in bridging the gap between AI and the physical world.