开源技术专家贾扬清宣布创办Intent Lab，担任创始人兼CEO[1]，并推出自主AI团队产品Fleet[1]。该产品旨在将意图转化为生产软件，具体涵盖需求理解、架构设计、编码开发、测试验证到持续运维的全流程[1]。
贾扬清此举系在英伟达收购Lepton AI仅一年后做出的重要职业决定[1]。英伟达于2025年4月以7亿美元完成收购[1]，并曾承诺于2026年开源Lepton核心平台[1]，但后来正式否决了该计划[1]。Fleet的联创由原Lepton AI团队成员组成，包括白俊杰和李响[1]。
Fleet已经取得初步成果。产品包括GLM 5.2推理引擎，其速度从102 tokens/s提升至647 tokens/s，提升幅度为6.3倍[1]；同时推出了一句话生成数据库引擎以及带形式化验证的文件系统[1]。
Jia Yangqing, a prominent figure in open-source development, has announced the founding of Intent Lab, where he serves as founder and CEO [1]. The startup introduces Fleet, an autonomous AI team product designed to transform requirements into production software by automating the full development lifecycle—from requirement analysis and architecture design through coding, testing, and continuous maintenance [1].
Jia's venture follows his departure from Nvidia approximately one year after the company acquired Lepton AI for $700 million in April 2025 [1]. According to the company background, Nvidia had committed to open-sourcing Lepton's core platform in 2026, but subsequently formally rescinded that commitment [1]. The founding team includes co-founders from the original Lepton AI, namely Bai Junjie and Li Xiang [1].
Fleet has already demonstrated significant technical progress [1]. Initial results include a GLM 5.2 inference engine that increased speed by 6.3 times, accelerating from 102 tokens per second to 647 tokens per second, as well as a one-line database engine generator and a file system incorporating formal verification [1].
