SDK 与客户端
一个基于 OpenAI 和 TypeSafe 的 Jev 终端界面，提供带透明决策报告的答案。
A polished OpenAI + TypeSafe Jev terminal interface for answers with transparent decision reports
from typesafe_sdk import Choice, Noul, RetryPolicy, Score, TypeSafeClient链接指向我们抓取当天的那个 commit，行号是准的。
下面是从这个项目源码里原样取出来的 question 组合。
取自 src/jev_system_one/jev.py:32
requires_uncertainty_notice是/否Should the answer explicitly disclose uncertainty or limits?
response_depth打分How much detail should the answer contain?
0 — brief: one direct answer with minimal context1 — standard: a concise answer with useful explanation2 — detailed: a thorough answer with structured reasoning or stepsimport { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
requires_uncertainty_notice: { type: 'boolean', instructions: 'Should the answer explicitly disclose uncertainty or limits?' },
response_depth: { type: 'score', instructions: 'How much detail should the answer contain?',
criteria: ['brief: one direct answer with minimal context', 'standard: a concise answer with useful explanation', 'detailed: a thorough answer with structured reasoning or steps'] },
},
});取自 src/jev_system_one/jev.py:107
requires_revision是/否Should OpenAI revise `openai_draft.answer` before it is shown?
import { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
requires_revision: { type: 'boolean', instructions: 'Should OpenAI revise `openai_draft.answer` before it is shown?' },
},
});