5 分钟上手
拿到 API key,用 cURL 或 SDK 跑通第一次 Jev 调用,并理解返回值结构。

第一步:在 Playground 里试
打开 Playground 并登录,把任意文本粘贴为 state:
Hi, I've been trying to connect my Stripe account for 3 days and it keeps failing.
I'm losing sales. Please help ASAP.
然后添加一个 Noul 问题:
{
"urgency": {
"type": "noul",
"instructions": "Does this message express urgency?"
}
}
你会立刻拿到一个 0 到 1 之间的数值。接近 1 表示模型认为答案是「是」。
Playground 的价值在于快速试错:混合 Noul、Choice、Score 三类问题,一次调用同时看到所有结果,确认问题措辞是否达到预期。
第二步:拿到 API key
在 dashboard 创建 key,然后设置环境变量:
export TYPESAFE_API_KEY="sk-..."
第三步:调用 API
所有模型都由同一个端点提供服务:
POST https://api.typesafe.ai/v1/systemone
Authorization: Bearer <API_KEY>
Content-Type: application/json
最小可运行的 cURL 示例:
curl -X POST https://api.typesafe.ai/v1/systemone \
-H "Authorization: Bearer $TYPESAFE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "jev-latest",
"state": "Hi, I have been trying to connect my Stripe account for 3 days and it keeps failing. I am losing sales. Please help ASAP.",
"questions": {
"department": {
"type": "choice",
"instructions": "Which team should handle this",
"criteria": {
"billing": "Payment or subscription issues",
"technical": "Bugs or integration problems",
"sales": "Pricing or account questions"
}
},
"frustration": {
"type": "score",
"instructions": "How frustrated the customer appears",
"criteria": [
"Calm, just stating facts",
"Frustrated but civil",
"Very angry, strong language"
]
},
"is_urgent": {
"type": "noul",
"instructions": "The message conveys urgency"
}
}
}'
第四步:用 SDK(推荐)
SDK 会默认从环境变量读取 TYPESAFE_API_KEY,并默认调用 jev-latest。
Python
pip install typesafe-sdk # 或 uv add typesafe-sdk
from typesafe_sdk import AsyncTypeSafeClient, Choice, Noul, Score
async def main() -> None:
async with AsyncTypeSafeClient() as client:
response = await client.system_one(
state={"document": "I was charged twice. Please fix this ASAP."},
questions={
"billing": Noul(instructions="Is this ticket about billing?"),
"tone": Choice(
instructions="What is the customer's tone?",
criteria={"calm": None, "frustrated": None, "angry": None},
),
"urgency": Score(
instructions="How urgent is this ticket?",
criteria=["can wait", "this week", "today"],
),
},
)
print(response.nouls["billing"].noul)
print(response.choices["tone"].choice)
print(response.scores["urgency"].score)
TypeScript / JavaScript
npm install @typesafe-ai/sdk # 需要 Node.js 20+
import { choice, TypeSafeClient } from "@typesafe-ai/sdk";
const client = new TypeSafeClient();
const response = await client.systemOne({
state: { document: "I was charged twice. Please fix this ASAP." },
questions: {
category: choice("What is this ticket about?", {
billing: null,
technical: null,
other: null,
}),
},
});
console.log(response.answers.category.choice);
TS SDK 的一大优势是答案类型由问题自动推导:传入的 questions 决定了返回值的类型,编译期就能发现字段名写错。
返回值长什么样
每个答案的类型由问题的 type 决定:
| 类型 | 返回字段 | 含义 |
|---|---|---|
choice |
choice / probabilities / confidence |
选中的选项、各选项概率分布、置信度 |
score |
score / legend / probabilities / confidence |
分数位置(可落在两档之间)、档位说明、概率分布、置信度 |
noul |
noul |
答案为「是」的概率,本身不含 confidence |
关键约束:答案永远落在你给定的选项内。模型返回的是你定义的选项上的概率分布,不会产生集合外的值,所以代码里不需要写解析器恢复语义。
常见坑
- 状态里的
state只读一次:模型先读一次 state,然后并行评估所有问题。所以趁早把多个问题打包进同一个请求,几乎没有额外延迟成本。 - 非文本输入要先转换:图片、音频、视频需要先转成文本或结构化字段再作为 state 传入。
- 超限返回 429:官方 SDK 默认按退避重试并尊重
retry-after头;直接调 HTTP API 需要自己实现。