At the recent Second CCF China Data Conference, Ant Digital officially announced the open-source release of its advanced data intelligence agent technology - Agentar SQL. This technology enables ordinary users to perform complex business data queries and analysis through simple everyday language, providing strong support for enterprises' digital transformation.

The first product released in this open-source initiative is a real-time text-to-structured query language (Text-to-SQL) framework, designed to help developers quickly build data query solutions and significantly improve the efficiency of interaction between text and databases. In the future, Ant Digital will continue to launch technologies covering database understanding, industry knowledge mining, and real-time multi-turn interactions, further enhancing data processing capabilities.

During the pilot operation at a leading city commercial bank, multiple tools of Agentar SQL demonstrated a query accuracy rate of over 92%, more than three times higher than traditional solutions. More impressively, Ant Digital's intelligent agent technology ranked among the top in the globally authoritative natural language to SQL evaluation benchmark BIRD-SQL, surpassing international giants like Google and becoming an industry leader.

image.png

BIRD-SQL's evaluation difficulty should not be underestimated, as its dataset covers 37 real-world scenarios including finance, power, and healthcare, with complex tasks and large data volumes, making it one of the most challenging NL2SQL tests globally. Research institutions estimate that by 2025, the global business intelligence market size will reach $47.48 billion, and the Chinese market will also expand rapidly, expected to reach $1.79 billion by 2028, with a compound annual growth rate of 12.7%.

Zhang Peng, the technical director of Ant Digital, pointed out at the conference that NL2SQL faces multiple challenges in practical applications, such as understanding the ambiguity of human speech, integrating industry-specific expertise, parsing complex database structures, and generating accurate SQL statements. Therefore, relying solely on simple models is far from sufficient.

Zhang Peng emphasized that to truly achieve industry-ready NL2SQL and data intelligence agent technologies, a complete capability system must be established. This includes in-depth understanding of databases, effective interaction between agents and users, and self-evolution capabilities. Ant Digital plans to continuously open-source more comprehensive capability modules in the future, further promoting the popularization of intelligent data analysis.

The open-source content of Agentar-Scale-SQL has been released on multiple platforms such as arXiv and GitHub, receiving widespread attention from developers.