Alibaba Cloud unveils Context Engine, ApsaraLakebase and Agent Context
At the 2026 Apsara Conference main technical forum, Alibaba Cloud proposed supporting agent deployment across three dimensions — Model, Harness and Context — and introduced a Context Engine that turns enterprise-wide data into real-time context agents can use. It also launched or upgraded the ApsaraLakebase open lakehouse storage, the AI-native big data service ADA, the enterprise agent-oriented Agent Context, plus PAI-RLS for reinforcement-learning training, the PAI-InferX inference platform and the AIDBS data-agent platform. Alibaba Cloud said new RDS PostgreSQL instances grew more than 400% quarter over quarter, with about 80% created by agents and roughly 50% of input/output traffic agent-driven; it also claims Agent Context can improve token efficiency by more than 3x and raise accuracy on complex knowledge tasks by up to 50 percentage points — all vendor-reported figures.
It shows major cloud vendors shifting competition from models themselves to the context and data foundation agents run on, where enterprise data governance and real-time supply may be the practical bottleneck for agent adoption.
在2026云栖大会技术主论坛上，阿里云提出面向Model、Harness、Context三个维度加持Agent的落地，其中决定Agent能力上限但最难标准化的一环是——面向Context，让Agent持续理解企业知识、业务状态和执行反馈。对此，阿里云智能集团首席技术官李飞飞进一步强调，围绕"存储—计算处理—上下文工程"，阿里云的Context Engine将企业全域数据转化为实时上下文，让Agent从"知道什么"走向"此刻应该做什么"。
基于这一理念，阿里云同步推出ApsaraLakebase、AI原生大数据服务ADA等产品的发布和重大升级。
上下文不再是会话级临时数据，而是企业新型数据资产
#Alibaba Cloud#Alibaba Cloud Context Engine#ApsaraLakebase#ADA#Agent Context#PAI-RLS#PAI-InferX#AIDBS#OpenLake#Kunlun Tech#Xiaohongshu