# JevModel

> Independent hosted playground, decision flows, batch processing and API for TypeSafe AI’s Jev decision model, with English and Chinese guides. JevModel is not affiliated with TypeSafe AI.

## What it does

Jev turns text or JSON state into typed judgments. Choice selects among named options, Score returns a possibly fractional value on an ordered rubric, and Noul returns a 0–1 probability for a true/false proposition. Applications must set their own action thresholds.

The Console manages saved decision flows, API keys, run history and token balance. A successful Decision Flow uses two daily free runs for the whole selected path; failed free flows refund both. With fewer than two free runs left, each visited question uses actual input tokens, and completed paid nodes remain charged if a later node fails. External actions belong to the calling application. Custom questions and criteria do not train or fine-tune a model.

Batch processing accepts CSV, TXT or JSONL, up to 2 MB and 1,000 rows, and exports results for your own evaluation. The service accepts text or JSON; provide existing captions or OCR text for image-related tasks. Probabilities are not correctness guarantees.

## Request shape

`POST /api/v1/systemone` accepts `{ "state": string | object | array, "questions": { "id": { "type": "choice" | "score" | "noul", "instructions": string, "criteria": ... } } }`. Choice criteria are a map of named options, Score criteria are an ordered array of levels, and Noul criteria contain `true` and `false` descriptions. Send 1–8 questions with stable lowercase IDs.

```json
{"state":"The customer was charged twice and requests a refund.","questions":{"escalate":{"type":"noul","instructions":"Should a human billing specialist review this request now?","criteria":{"true":"A specialist should review it now.","false":"No immediate review is needed."}}}}
```

Create an API key after email verification and send it as `Authorization: Bearer <key>`. The playground, batch processing, flows and API share ten daily free units per UTC day, subject to site capacity: a regular run uses one unit and a Decision Flow uses two. Signup does not grant tokens. Additional usage draws from subscription tokens that refresh monthly or one-time token packs that never expire. Console and account pages require sign-in and are excluded from indexing.

## Pages

- [Runnable API examples on GitHub](https://github.com/SouthCat196/jevmodel-examples): Independent JevModel Node.js, Python and curl examples for Choice, Score and Noul. Offline fixtures by default; --live explicitly opts into an API call. MIT example code, not Jev model weights or an official TypeSafe SDK.
- [Home](https://jevmodel.app/)
- [中文首页](https://jevmodel.app/zh)
- [Playground overview](https://jevmodel.app/workbench)
- [Batch processing](https://jevmodel.app/batch)
- [Decision flows](https://jevmodel.app/#flows)
- [Use cases](https://jevmodel.app/use-cases)
- [Guides](https://jevmodel.app/guides)
- [API guide](https://jevmodel.app/guides/api)
- [Usage rules](https://jevmodel.app/pricing)
- [Full guide text](https://jevmodel.app/llms-full.txt)

## Guides

- [Structured Decision Models for Autonomous Agents](https://jevmodel.app/guides/structured-decision-models): Design bounded agent decisions with Choice, Score and Noul. Compare hosted Jev and visual Valen, connect Codex, and evaluate errors, latency and cost.
- [自主 Agent 的结构化决策模型指南](https://jevmodel.app/zh/guides/structured-decision-models): 如何把 Agent 的下一步动作变成可评估的结构化判断？了解 Choice、Score、Noul，比较托管 Jev 与视觉 Valen，接入 Codex，并评估错误、延迟和成本。
- [Valen vs Jev in 2026: Which Decision Model Fits Your Agent?](https://jevmodel.app/guides/valen-vs-jev): Compare Valen and Jev for agent decisions: visual inputs, open weights, published Sokoban benchmarks, self-hosting steps and the limits of each approach.
- [Valen 与 Jev 对比（2026）：你的 Agent 该选哪种决策模型？](https://jevmodel.app/zh/guides/valen-vs-jev): Valen（万澜）与 Jev 有什么关系，Agent 应该怎么选？对比视觉输入、开放权重、作者公布的推箱子评测、自行部署步骤及成本，区分演示成绩与实际生产表现。
- [What is JevModel, and how is it related to Jev?](https://jevmodel.app/guides/what-is-jevmodel): JevModel is an independent hosted Jev playground and API for text decisions, with visual flows and browser batch processing. TypeSafe AI develops Jev.
- [JevModel 是什么？与 Jev 有什么关系？](https://jevmodel.app/zh/guides/what-is-jevmodel): JevModel 是独立运营的 Jev 文本判断工作台与 API，提供可视化流程和浏览器批处理；Jev 模型由 TypeSafe AI 开发。
- [How to classify text with the JevModel API](https://jevmodel.app/guides/text-classification-api): Classify text into your own labels with a Choice request. See the input, illustrative response, error handling and human-review path.
- [如何用 JevModel API 分类文本](https://jevmodel.app/zh/guides/text-classification-api): 用 Choice 请求把文本归入自定义标签；查看输入、示意响应、错误处理与人工复核路径。
- [Jev probability, confidence and human-review thresholds](https://jevmodel.app/guides/confidence-and-abstention): Understand Choice confidence, Noul probability and Score rubrics. Evaluate thresholds on labeled data before automating a decision.
- [Jev 概率、置信度与人工复核阈值](https://jevmodel.app/zh/guides/confidence-and-abstention): 区分 Choice 置信度、Noul 概率与 Score 量表；在自动化决策前，用标注数据评估阈值。
- [Jev vs embeddings and zero-shot classifiers](https://jevmodel.app/guides/jev-vs-embeddings): Compare typed decisions, vector similarity and zero-shot label scoring for text classification. Choose using your own data and deployment constraints.
- [Jev、Embeddings 与零样本分类器如何选型](https://jevmodel.app/zh/guides/jev-vs-embeddings): 比较类型化判断、向量相似度与零样本标签评分；依据实际数据和部署约束选择文本分类方案。
- [Clef vs Jev: choosing a decision model for your workflow](https://jevmodel.app/guides/clef-vs-jev): Compare Cloudflare Clef with hosted Jev by input needs, API boundaries and deployment. Use a task-specific evaluation before changing your decision layer.
- [Clef vs Jev：如何为实际工作流选择决策模型](https://jevmodel.app/zh/guides/clef-vs-jev): 按输入需求、API 边界和部署方式比较 Cloudflare Clef 与托管 Jev；区分厂商成绩与实际工作流表现，了解替换前应验证哪些条件。
- [Cloudflare Clef and Clef-flash: a practical integration guide](https://jevmodel.app/guides/cloudflare-clef): Learn how Cloudflare Clef takes typed questions, where to call its API, and what to validate before using its answers in an agent or review workflow.
- [Cloudflare Clef 与 Clef-flash：接入与验证指南](https://jevmodel.app/zh/guides/cloudflare-clef): 了解 Cloudflare Clef 的类型化问题、真实 API 路径、输入限制和部署选择；在将答案接入 Agent 或复核流程前验证协议与安全边界。
- [Strands Decider 2B: what it does and how to compare it with Jev](https://jevmodel.app/guides/strands-decider-2b): Understand AWS Strands Labs’ experimental open decision model, checkpoint versions, local serving and a practical Jev evaluation plan.
- [Strands Decider 2B：能力、部署与 Jev 对比](https://jevmodel.app/zh/guides/strands-decider-2b): 了解 AWS Strands Labs 的实验性开放决策模型、checkpoint 版本、本地服务与 Jev 对照评估方法。
- [OpenAI Decisions API vs Jev: contracts, costs and migration](https://jevmodel.app/guides/openai-decisions-api-vs-jev): Compare OpenAI’s public-beta Decisions API with Jev: predicate, choice and score outputs, JSON shapes, refusal handling and a migration checklist.
- [OpenAI Decisions API 与 Jev：接口、成本与迁移](https://jevmodel.app/zh/guides/openai-decisions-api-vs-jev): 对比 OpenAI 公开测试的 Decisions API 与 Jev：predicate、choice、score 输出、JSON 结构、拒答处理与迁移检查。
- [Use Jev in Codex with the Official TypeSafe Skill](https://jevmodel.app/guides/codex-jev): Install the official TypeSafe skill in Codex, configure TYPESAFE_API_KEY, and ask Jev to select a next action with a verified API call.
- [Codex 接入 Jev：官方 TypeSafe Skill 教程](https://jevmodel.app/zh/guides/codex-jev): 在 Codex 安装官方 TypeSafe Skill、配置 TYPESAFE_API_KEY，并通过真实 Jev API 调用选择下一步动作，包含完整提示词、验证和排错步骤。
- [What is the Jev model?](https://jevmodel.app/guides/what-is-jev): Understand TypeSafe AI’s Jev decision model, its Choice, Score and Noul outputs, and when to use it.
- [Jev 模型是什么？](https://jevmodel.app/zh/guides/what-is-jev): 了解 TypeSafe AI 的 Jev 判断模型、Choice、Score、Noul 输出及适用场景。
- [Your first Jev decision](https://jevmodel.app/guides/getting-started): Try a support-routing example, read its Choice probabilities, and find your next steps for the JevModel API.
- [第一次运行 Jev 判断](https://jevmodel.app/zh/guides/getting-started): 用客服分流示例运行 Jev 判断、阅读 Choice 概率，并了解后续 API 接入方式。
- [JevModel API: request and response](https://jevmodel.app/guides/api): Use a JevModel API key to send typed Jev questions to /api/v1/systemone.
- [JevModel API：请求与响应](https://jevmodel.app/zh/guides/api): 使用 JevModel API 密钥，通过 /api/v1/systemone 发送类型化判断问题。
- [Choice, Score, and Noul](https://jevmodel.app/guides/question-types): Pick the answer type that matches the decision your code will make.
- [Choice、Score 与 Noul](https://jevmodel.app/zh/guides/question-types): 按照代码需要的结果，选择合适的问题类型。
- [Where typed decisions help](https://jevmodel.app/guides/use-cases): Five starter workflows, from support routing to agent control.
- [类型化判断适合哪些场景](https://jevmodel.app/zh/guides/use-cases): 从客服分流到 Agent 控制，了解五个入门流程。
- [Route customer support with Jev](https://jevmodel.app/guides/support-triage): Classify a message and decide when a person should take over.
- [用 Jev 进行客服分流](https://jevmodel.app/zh/guides/support-triage): 分类客户消息，并判断何时交给人工。
- [Use Jev for review queues](https://jevmodel.app/guides/content-moderation): Triage posts for review without treating a probability as a final verdict.
- [用 Jev 辅助内容审核](https://jevmodel.app/zh/guides/content-moderation): 用概率安排复核，不把概率当作最终裁决。
- [Score leads on a useful rubric](https://jevmodel.app/guides/lead-scoring): Make sales prioritization easier to test and revise.
- [用清晰量表评估线索](https://jevmodel.app/zh/guides/lead-scoring): 让销售优先级更容易检验与修订。
- [Check a claim against evidence](https://jevmodel.app/guides/evidence-checking): Use a narrow proposition to spot unsupported statements.
- [核验主张与证据](https://jevmodel.app/zh/guides/evidence-checking): 用范围明确的命题识别缺乏支持的表述。
- [Route an agent’s next step](https://jevmodel.app/guides/agent-routing): Choose between search, calculation, response, and clarification.
- [为 Agent 选择下一步](https://jevmodel.app/zh/guides/agent-routing): 在检索、计算、回答与追问之间选择。
- [Add a decision gate before tool use](https://jevmodel.app/guides/tool-guardrails): Combine a model signal with deterministic permission checks.
- [在工具调用前增加判断关口](https://jevmodel.app/zh/guides/tool-guardrails): 将模型信号与确定性的权限校验结合。
- [Choose what context to keep](https://jevmodel.app/guides/context-compaction): Use typed questions before trimming a long agent history.
- [判断应保留哪些上下文](https://jevmodel.app/zh/guides/context-compaction): 在缩减 Agent 长历史前使用类型化问题。
- [Jev vs GPT and Claude: choosing a decision layer](https://jevmodel.app/guides/jev-vs-llm): Compare Jev typed probabilities with GPT and Claude structured outputs. Choose routing, generation or a hybrid workflow using your own test cases.
- [Jev vs GPT 与 Claude：如何选择判断层](https://jevmodel.app/zh/guides/jev-vs-llm): 对比 Jev 类型化概率与 GPT、Claude 的结构化输出；通过真实测试案例选择路由判断、内容生成或混合工作流。
- [Jev or rules?](https://jevmodel.app/guides/jev-vs-rules): Decide when a model helps and when ordinary code is enough.
- [Jev 与规则引擎怎么选？](https://jevmodel.app/zh/guides/jev-vs-rules): 分清模型何时有用、普通代码何时足够。
- [Evaluate a Jev workflow](https://jevmodel.app/guides/evaluation): Test labels, uncertainty, and real outcomes before automation.
- [评估 Jev 工作流](https://jevmodel.app/zh/guides/evaluation): 在自动化前测试标签、不确定性和真实结果。
- [Private runs and history](https://jevmodel.app/guides/privacy-and-history): What the JevModel workbench stores and how to remove it.
- [运行记录与隐私](https://jevmodel.app/zh/guides/privacy-and-history): 了解 JevModel 工作台保存什么，以及如何删除。
- [Ways to access Jev](https://jevmodel.app/guides/access-paths): Choose a provider path and keep credentials on your server.
- [Jev 的接入方式](https://jevmodel.app/zh/guides/access-paths): 选择模型提供方，并将凭据留在服务端。
- [Is Jev open source?](https://jevmodel.app/guides/is-jev-open-source): Separate the hosted Jev model from independent open alternatives.
- [Jev 是开源的吗？](https://jevmodel.app/zh/guides/is-jev-open-source): 区分托管 Jev 模型与独立开源替代方案。
- [What is Laya?](https://jevmodel.app/guides/what-is-laya): An open decision model with a Jev-shaped task, from a different team.
- [Laya 是什么？](https://jevmodel.app/zh/guides/what-is-laya): 了解由另一团队开发、同样处理类型化判断的开放模型。
- [Jev vs Laya](https://jevmodel.app/guides/jev-vs-laya): Hosted Jev access and open local deployment solve different constraints.
- [Jev 与 Laya 怎么选？](https://jevmodel.app/zh/guides/jev-vs-laya): 比较 Jev 的托管接入与 Laya 的本地部署边界。
- [Jev vs OpenJev](https://jevmodel.app/guides/jev-vs-openjev): API familiarity is useful, but runtime and model behavior still differ.
- [Jev 与 OpenJev 怎么选？](https://jevmodel.app/zh/guides/jev-vs-openjev): 接口相似有助接入，但运行方式与模型行为仍然不同。
- [Jev vs djev](https://jevmodel.app/guides/jev-vs-djev): Compare text decisions with a DiffusionGemma-based multimodal approach.
- [Jev 与 djev 怎么选？](https://jevmodel.app/zh/guides/jev-vs-djev): 比较文本判断与基于 DiffusionGemma 的多模态方案。
- [SemIf vs Jev: Benchmarks, Calibration, and Deployment](https://jevmodel.app/guides/jev-vs-semif): Compare SemIf vs Jev using primary sources: option scoring, benchmark limits, CPU and GPU deployment, calibration, and a practical workflow evaluation plan.
- [SemIf vs Jev：评测、概率校准与部署选型](https://jevmodel.app/zh/guides/jev-vs-semif): 依据官方文档与原始仓库比较 SemIf vs Jev：解释公开评测的边界、CPU/GPU 与 Apple Silicon 部署、概率校准、数据路径和总成本，并给出客服分流示例与选型检查方法。
- [Jev vs Jev-Omni: text and multimodal decisions](https://jevmodel.app/guides/jev-omni): Jev-Omni is an independent 12B model for text, image, audio, and video decisions. Compare its inputs and hardware needs with hosted Jev.
- [Jev 与 Jev-Omni：文本与多模态判断](https://jevmodel.app/zh/guides/jev-omni): Jev-Omni 是独立的 12B 文本、图像、音频和视频判断模型；了解它与托管 Jev 的输入及硬件区别。
- [JevModel pricing and free Jev runs](https://jevmodel.app/guides/pricing-and-free-tier): Understand verified daily Jev runs, Flow usage, input-token billing and service packs. Choose a plan after measuring your own workload.
- [JevModel 价格与 Jev 免费额度](https://jevmodel.app/zh/guides/pricing-and-free-tier): 了解邮箱验证后的 Jev 每日免费额度、Flow 用量、输入 Token 计费与本站套餐；先测实际工作量，再选择方案。
- [AI Decision Flows with Jev and human review](https://jevmodel.app/guides/decision-flows): Build a Jev decision flow for support routing, add an uncertain-answer branch and call the saved graph through the API. Learn limits and billing.
- [Jev AI 决策流与人工复核分支](https://jevmodel.app/zh/guides/decision-flows): 搭建 Jev 客服决策流、添加不确定结果人工复核分支，并通过 API 调用已保存流程；了解路径、限制与计费。
- [Batch text classification with the JevModel API](https://jevmodel.app/guides/batch-text-classification-api): Classify CSV or JSONL rows with Jev Choice questions. Reuse the normal API, preserve row IDs and export results without inventing a batch endpoint.
- [Jev API 批量文本分类：CSV 与 JSONL](https://jevmodel.app/zh/guides/batch-text-classification-api): 用 Jev Choice 分类 CSV 或 JSONL 行，复用普通 API、保留行 ID 并导出结果；了解大小限制、失败处理与计费。