Agent 工具

jev-use

@shitianfang8JavaScriptMIT更新于 2026-09-19可当场跑

Claude Code、Codex 和 pi 插件,将无需文本输出的 agent 步骤交给 Jev(TypeSafe 的判断模型)处理,延迟约 230 毫秒,成本约 0.02 美元每千次判断,支持类型化回退到 LLM。

英文原文

Claude Code / Codex / pi plugin that hands agent steps needing no text output to Jev (TypeSafe's judgment model) — measured p50 ~230 ms and ~$0.02 per 1,000 judgments, with typed escalation back to the LLM

shitianfang/jev-use

它在哪儿调用了 Jev

* POST https://api.typesafe.ai/v1/systemone, Bearer TYPESAFE_API_KEY.

src/backends/typesafe.ts:3

链接指向我们抓取当天的那个 commit,行号是准的。

它问 Jev 的问题

下面是从这个项目源码里原样取出来的 question 组合。

取自 test/backends.test.ts:94

  1. is_bug是/否

    Is the customer reporting a software defect?

  2. team选项

    Which team should own this ticket?

    • frontendRendering issues
    • paymentsCheckout issues
  3. urgency打分

    How urgent is this ticket?

    • 0Can wait
    • 1This week
    • 2Blocking revenue

用你自己的内容跑一遍

代码
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    is_bug: { type: 'boolean', instructions: 'Is the customer reporting a software defect?' },
    team: { type: 'choice', instructions: 'Which team should own this ticket?',
      criteria: { frontend: 'Rendering issues', payments: 'Checkout issues' } },
    urgency: { type: 'score', instructions: 'How urgent is this ticket?',
      criteria: ['Can wait', 'This week', 'Blocking revenue'] },
  },
});

同类的其他项目