jevIMAGE
Grev harness：Grok Build 接 Jev 做代码库检索
作者用 Jev 在约 2 秒内给上千文件排序，让模型直接基于证据回答，探索阶段提速约 40%、token 省 80%。
Nadir
@NadirMw2026.09.21收录原推发布于 2026.09.20
Nadir @NadirMw原推发布于 2026.09.20
Built Grev harness: Grok Build x Jev. Sign in with your ChatGPT sub to run GPT-6 Astra. ~40% faster >80% fewer tokens on codebase exploration. The LLM says what it needs in plain words. Jev ranks ~1000 files in ~2s. The model answers straight from evidence. Exploration only codegen untouched. Still wiring autocompact for more savings. Interested if I drop this on GitHub?
x.com/nadirmw/status/2101421434899701856采集于 2026.09.21 10:31