Integrating the Jev judgment model with Codex
Juejin author 二哥呀 demonstrates how to install TypeSafe AI’s Jev Skill in Codex, configure an API key, and require the agent to call Jev’s Choice API before taking an action. In the shown example, Jev 1.13.0 selected web search for an Nvidia stock-price question with 0.99 confidence, after which Codex executed the selected action. The author also says Jev works with Claude Code and PI, while speed and cost comparisons come from TypeSafe AI rather than independent testing.
The workflow offers coding agents a concrete way to separate decision selection from action execution, but its broader compatibility and efficiency advantages remain unverified.
大家好，我是二哥呀。
昨天在王二讲 Agent 的视频里分享了 Jev，播放直接爆了。
看来大家对 Jev 的热情是真的高。
简单介绍下，Jev 是一种判断模型（Judgment Model），我们输入一个状态和几个选项，它告诉我们该选哪个，置信度多少。
就这么简单。
下图是我用Codex配合Jev填写一个千问的表单，真的很无敌，速度超级快，效果还超级好。