Huawei openJiuwen 推出 TTSE 智能体框架：将环境事实与条件提示分离
Huawei's openJiuwen Unveils TTSE, an Agent Framework That Splits Environment Facts From Tips
华为 openJiuwen 开源了 TTSE，把智能体经验分成环境事实和条件提示两轨管理，在 ALFWorld 跑到 97.8%，还减少了办公任务的工具调用。做智能体长期记忆和经验积累的可以看看源码，不过数据是自家评测，先自己跑一遍再下结论。
Huawei 开源团队 openJiuwen 发布 TTSE（Two-Track Self-Evolution）智能体经验框架，把智能体积累的经验拆成两类：环境事实（Env Facts）与条件提示（Tips）。环境事实记录任务环境的稳定属性，条件提示则绑定触发条件，只在特定情境下生效，避免经验互相干扰。据团队自报数据，TTSE 在 ALFWorld 基准上取得 97.8% 的成绩，并在办公任务演示中减少了工具调用次数。
技术上，TTSE 的双轨设计解决了单一经验库常见的“规则过拟合”问题：事实类经验可跨任务复用，提示类经验则通过条件匹配按需检索，降低了上下文膨胀和错误经验被反复执行的风险。
需要注意的是，ALFWorld 97.8% 与工具调用削减数据均来自 openJiuwen 自己的评测，尚无第三方复现。与同类经验自进化方案（如 Voyager、Reflexion）相比，TTSE 的差异点在于显式区分经验类型。对做智能体经验管理的开发者，该框架已随 openJiuwen 开源。
Huawei's openJiuwen Unveils TTSE, an Agent Framework That Splits Environment Facts From Tips
openJiuwen's TTSE sorts agent experience into environment facts and conditional tips. The team reports 97.8% on ALFWorld and cuts office-task tool calls in demos; results are its own.
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include openJiuwen, Huawei, 智能体, ALFWorld.
Primary source verification: published directly by pandaily with first-party authority.