On November 18, during the 2025 OceanBase Annual Conference, OceanBase launched and open-sourced its first AI database, OceanBase SeekDB (short for SeekDB). Developers can quickly build AI applications such as knowledge bases and agents with just three lines of code, easily handling multi-modal data retrieval at the scale of billions, truly achieving "out-of-the-box" AI data foundation.

The product supports unified hybrid search for vectors, full-text, scalars, and spatial geographic data, deeply integrating AI reasoning with data processing, and is compatible with over 30 mainstream AI frameworks such as Hugging Face and LangChain. This product realizes a new function for databases to move from traditional "business support systems" to "AI-native data entry points," and is also a result of OceanBase's "Data×AI" strategy since its launch. CEO Yang Bing said: "We hope OceanBase can explore a paradigm shift for databases in the AI era."

5683f979c806582cf275eb34e7bd5d4.png

Yang Bing believes that the real bottleneck of AI is not in the model, but in the data. Especially in high-sensitive scenarios like finance and government affairs, AI needs to perform real-time inference within milliseconds and safely integrate private data. However, traditional architectures rely on multi-system data pipelines, which are complex, inefficient, and prone to permission confusion and latency risks.

"SeekDB is not just a combination of traditional database functions, but an AI-native database restructured for the AI era," Yang Bing said. "It inherits OceanBase's code and design philosophy, being lighter and more agile, with the goal of becoming the 'real-time entry layer' for large models and private data fusion computing. We hope to work with developers to accelerate iteration and boldly innovate in areas such as hybrid search and multi-modal fusion."

Gartner predicts that by 2028, database spending supporting generative AI will reach $218 billion, accounting for 74% of the market. However, MIT research shows that over 95% of enterprise AI projects fail to be implemented due to fragmented multi-modal data, long system pipelines, and complex permission management. SeekDB brings three core breakthroughs:

First, AI-native hybrid search capabilities. SeekDB supports the integration of vector retrieval, full-text search, and scalar filtering in a single query, using a "coarse ranking + fine ranking" multi-stage retrieval mechanism to improve accuracy while maintaining low latency. Relying on a mature transaction engine, it supports real-time writing and ACID consistency, and is compatible with the MySQL ecosystem. Additionally, SeekDB supports unified storage and retrieval of multi-modal data such as scalars, vectors, text, JSON, and GIS. For example, in anti-fraud scenarios, it can directly query "transactions over 50,000 yuan in the last 7 days, abnormal location, and similar behavior to historical fraud samples" without cross-system calls, balancing performance and security.

Second, extremely simple deployment and out-of-the-box usability. SeekDB requires as little as 1 CPU core and 2GB of memory, supports pip install for one-click installation and instant startup, and is compatible with embedded and client/server dual deployment modes, making it easy to integrate into smart agents, development toolchains, or local applications, significantly lowering the engineering barriers for AI applications.

Third, greater developer-friendliness. SeekDB is globally open-sourced under the Apache 2.0 license, allowing developers to freely use, modify, and expand it. The product fully supports over 30 AI frameworks such as HuggingFace, Dify, and LangChain, as well as the MCP large model protocol, seamlessly integrating into the AI ecosystem; it also provides SQL and Python SDKs, adapting to different development habits. At the same time, the PowerRAG intelligent document parsing framework and PowerMem hierarchical memory architecture were also open-sourced. The latter achieved a top score of 73.70 on the LOCOMO Benchmark, reducing token consumption by 96%, significantly saving inference costs.

As a key part of OceanBase's "Data×AI" strategy, SeekDB can be used independently or smoothly integrated into the newly released OceanBase 4.4 integrated version. This version first integrates TP, AP, and AI capabilities into a single kernel, combining distributed expansion, multi-cloud deployment, and financial-grade high availability, helping enterprises avoid the risk of architectural restructuring later. The commercial LTS version will be released on February 2, 2026.

Currently, OceanBase's hybrid search capabilities have been successfully implemented in multiple industries, fully verifying their technical value: China Unicom built a unified AI knowledge base based on hybrid search, effectively solving the problems of private document permission management and efficient retrieval; Ant Baibaoxiang achieved real-time online search for agents through hybrid search, significantly improving the accuracy and response efficiency of information retrieval.

"This is not just a technical product, but a shift in development paradigms," Yang Bing said. "Traditional databases only 'store' data, while SeekDB can 'understand' data semantics. Hybrid search is the key turning point for AI-native databases." Over the past 15 years, OceanBase has honed its engineering capabilities in extreme scenarios like Double 11, which are now transforming into foundational advantages in the AI era, continuously breaking through in areas such as AI-native hybrid search, multi-modal fusion, TP/AP/AI integration, and multi-cloud-native.

Since its independent development by Alibaba Group in 2010, OceanBase has served more than 4,000 enterprises worldwide, covering key sectors such as finance, government and enterprise, energy, communications, retail, manufacturing, and the internet. Its cloud service, OB Cloud, is the only database product in the world that simultaneously supports seven major cloud vendors including Alibaba Cloud, Huawei Cloud, Tencent Cloud, Baidu Intelligent Cloud, AWS, GCP, and Azure. Its business spans 16 countries and regions, 60 locations, and 240 availability zones around the globe.