把 Jev 当 RAG 重排序器，给上下文工程省钱

Viv @Vtrivedy10原推发布于 2026.10.01

在 X 上打开原推

Context Engineering is still the most important job of agent & harness engineers Our goal: efficiently route the highest quality context into the LLM computational boundary (ie. the context window) a big part of this is retrieval! ppl are using Jev for RAG (as a reranker) 👀 targetedly inserting cheap intelligence into context gathering steps matters can give you great perf/cost boosts collecting some previous diagrams below on - Harnesses as context routers - Jev for RAG

这条是引用着另一条写的：x.com/erikkaum/status/2103169247102812334