从发布线程进入 Jev 技术博客
Diogo 在发布线程末尾给出官方技术博客与早期访问入口,适合继续核对模型原理和产品说明。
开发者分享使用 TypeSafe JEV 与 Sanity 的演示:查询路径不含 LLM,智能体尽可能跳过检索,无法跳过时也仅单向执行,文本在摄入阶段已生成,智能体只负责选择答案。
I just built a demo that uses @typesafeai JEV and @sanity_io and it's incredible. No LLM in the query path. The "agent" skips retrieval entirely when it can, and when it can't, it's one-way (no roundtrips). The prose was already written at ingest, so the agent selects an answer