System One

System One 模型为软件做出快速、结构化的决策。Jev 是 TypeSafe 的旗舰模型，也是首个 System One 模型。

System One models are a class of AI models built to make fast, structured decisions that software can use directly. A System One model evaluates a state and returns typed answers and probabilities.

System One 模型是一类 AI 模型，专门用于做出软件可以直接使用的快速、结构化决策。一个 System One 模型会评估一个 state（上下文状态），并返回类型化的答案与概率。

Jev is TypeSafe's flagship model and the first System One model.

Jev 是 TypeSafe 的旗舰模型，也是首个 System One 模型。

Like an LLM, a System One model understands natural-language input. It returns typed decisions and probabilities rather than generated text.

和 LLM 一样，System One 模型能够理解自然语言输入。它返回的是类型化的决策与概率，而不是生成的文本。

与 LLM 的区别

System One models are trained for calibrated decisions: their probabilities are optimized against outcomes to reflect uncertainty. Calibration is measured across groups of predictions; it does not guarantee that an individual answer is correct.

System One 模型为校准过的决策而训练：其概率会针对真实结果进行优化，以反映不确定性。校准是在一组预测之上进行衡量的，并不能保证单个答案一定正确。

System One models do not write replies, produce code, or generate explanations of their reasoning. You define the possible answers through primitives:

System One 模型不会撰写回复、生成代码，也不会对自身推理过程给出解释。你需要通过 primitives（题型） 来定义可能的答案：

These are illustrative configurations and values. The primitive pages describe the available configuration options and full response fields.

以上仅为示意性的配置和取值。各 primitive 页面详细描述了可用的配置选项以及完整的响应字段。

Read the AI primer to learn how System One models work and how they are trained.

若想了解 System One 模型的工作原理及训练方式，请阅读 AI 入门（AI primer）。

在更大工作流中做快速判断

For a refund request, your application can:

对于一次退款请求，你的应用可以：

1. Build a state containing the customer's message, the relevant transactions, and the refund policy.

2. 构建一个 state，其中包含客户消息、相关交易记录以及退款政策。

3. Ask independent questions together: whether a refund was requested, whether the evidence indicates a duplicate charge, and whether the policy supports a refund.

4. 一次性提出多个相互独立的问题：是否提出了退款请求、证据是否表明存在重复扣款、政策是否支持退款。

5. Combine the answers with deterministic checks in code, then route the case for action or review.

6. 在代码中把这些答案与确定性的检查组合起来，再将案件路由到相应的处理动作或人工复核。

Once you have seen the primitives in action, you can combine them into a larger system. Because System One models return typed, constrained outputs rather than free-form text, your code can inspect and combine its answers into predictable workflows. See How to build with TypeSafe for the full workflow.

在见识过 primitives 的实际运行效果之后，你便可以把它们组合成更大的系统。由于 System One 模型返回的是类型化、受约束的输出，而不是自由格式的文本，你的代码可以检视并组合这些答案，形成可预测的工作流。完整工作流请参阅 How to build with TypeSafe。

Answers to some question types also include confidence (each type's page says which), so you can decide when to act and when to escalate to a person or a reasoning model.

部分问题类型的答案还会附带 confidence（置信度）（具体包含哪些字段见各类型页面），便于你决定何时直接采取行动、何时升级给人工或推理模型。

调用 System One 模型

Call a System One model through one of our client SDKs or POST /v1/systemone in the HTTP API. The model field selects which model handles the request. The examples in these docs use jev-latest, which is also the SDK default. See Models for the available models, their prices, and their aliases.

你可以通过我们的 客户端 SDK 之一，或 HTTP API 中的 POST /v1/systemone 来调用 System One 模型。model 字段用于选择处理本次请求的模型。本文档示例使用的是 jev-latest，这同时也是 SDK 的默认模型。可用模型、价格以及别名请参阅 Models。

Start with State to prepare the input and Primitives (Questions) to explore the types of questions you can ask.

你可以从 State 开始学习如何准备输入；再阅读 Primitives (Questions)，了解你可以提出的问题类型。

This documentation is built and hosted on Mintlify, a developer documentation platform.

本文档由面向开发者的文档平台 Mintlify 构建并托管。