# Jev 101

> Independent bilingual learning guide to Jev, with a source-backed resource directory. Not affiliated with TypeSafe.

## Important

Documentation reviewed 2026-09-19. Demos are invented fixtures, not live model results. Project listings are snapshots, not endorsements. This static site has no inference API or MCP server.

## Learning guide

- [Jev 是什么？先看懂它的工作](https://jev101.com/markdown/zh/what-is-jev.md): 不是再添一个聊天窗口，而是给软件增加一个能理解文字的判断环节。
- [和大模型、规则、向量检索有什么不同？](https://jev101.com/markdown/zh/jev-vs-llms.md): 用同一个客服任务看输入、输出与分工，而不是拿宣传倍数比高低。
- [Choice、Score、Noul：怎么选？](https://jev101.com/markdown/zh/three-primitives.md): 分类、分档评分、判断是否存在，分别使用合适的提问方式。
- [第一个任务，选小一点](https://jev101.com/markdown/zh/choose-first-task.md): 先做可复核、低风险的判断，不从全自动经营或复杂代理开始。
- [从哪里获取 Jev API？](https://jev101.com/markdown/zh/api-channels.md): 区分 TypeSafe 直连与服务商接入：密钥、模型名和请求格式不能混用。
- [获取密钥，并安全保存](https://jev101.com/markdown/zh/get-api-key.md): 从官方控制台开始；学习站不接收、不保存你的 API Key。
- [完成第一次调用：HTTP、Python 与 JavaScript](https://jev101.com/markdown/zh/first-api-call.md): 从一条脱敏文本得到一个真实的类型化答案，保留模型和用量信息。
- [让编程 Agent 帮你接入 Jev](https://jev101.com/markdown/zh/coding-agent-skill.md): Skill 教 Agent 怎么用接口；它不替你提供密钥，也不等于强制执行流程。
- [准备资料，问对问题](https://jev101.com/markdown/zh/state-and-questions.md): 把复杂需求拆成可检查的小判断，减少噪声和隐含条件。
- [实战一：售后邮件分流](https://jev101.com/markdown/zh/support-routing.md): 从一封邮件到一个待确认标签，不直接触发退款或对外回复。
- [实战二：筛选 RAG 参考资料](https://jev101.com/markdown/zh/rag-reranking.md): 先检索候选，再判断相关性；不把相似度、相关性与事实支持混成一件事。
- [实战三：在 n8n 里接入 Jev](https://jev101.com/markdown/zh/n8n-workflow.md): 先用通用 HTTP Request 节点走通判断链，不依赖未经核对的社区插件。
- [实战四：选择候选字段，再由代码校验](https://jev101.com/markdown/zh/structured-extraction.md): 先找候选日期、金额或编号；让判断与解析、运算各司其职。
- [看懂概率、置信度与人工回退](https://jev101.com/markdown/zh/confidence-and-fallback.md): 类型正确不等于结论正确；门槛应由真实标注数据决定。
- [中国用户必读：先验证中文效果](https://jev101.com/markdown/zh/evaluate-chinese.md): 同一个产品支持中文，不代表它对你的行业语言已经可靠。
- [算清费用，测对速度](https://jev101.com/markdown/zh/cost-and-latency.md): 把调用量、输入规模与失败重试算进去，别只看模型的宣传数字。
- [上线前：权限、数据与执行关卡](https://jev101.com/markdown/zh/production-checklist.md): 将“模型说可以”与“系统允许执行”严格分开。
- [常见问题：报错、选错与接不通](https://jev101.com/markdown/zh/troubleshooting.md): 先分清账户、接口、模型判断与执行流程是哪一层出了问题。
- [术语速查：不被缩写绕晕](https://jev101.com/markdown/zh/glossary.md): 把模型、接口、插件与工作流中的名词放回各自位置。
- [版本、来源与更新规则](https://jev101.com/markdown/zh/versions-and-sources.md): 清楚区分文档核对、代码离线测试与实际服务验证。
- [What is Jev? Start with one decision](https://jev101.com/markdown/en/what-is-jev.md): A decision component for software, not another chat window.
- [Jev vs LLMs, rules and retrieval](https://jev101.com/markdown/en/jev-vs-llms.md): Compare responsibilities on one task—not unsupported speed claims.
- [Choice, Score and Noul: which one?](https://jev101.com/markdown/en/three-primitives.md): Choose between a category, an ordered scale and a probability of yes.
- [Choose a small first task](https://jev101.com/markdown/en/choose-first-task.md): Start with a reviewable, low-risk decision.
- [Where can you access the Jev API?](https://jev101.com/markdown/en/api-channels.md): Separate direct access from providers: keys, model IDs and schemas differ.
- [Get an API key and keep it private](https://jev101.com/markdown/en/get-api-key.md): Start at the official console. This guide never collects your key.
- [Make your first call: HTTP, Python and JavaScript](https://jev101.com/markdown/en/first-api-call.md): Turn one redacted text into a typed answer and inspect usage.
- [Build with a coding agent and the official skill](https://jev101.com/markdown/en/coding-agent-skill.md): A skill teaches the interface. It is not a key or an enforcement engine.
- [Prepare state and ask precise questions](https://jev101.com/markdown/en/state-and-questions.md): Turn a broad request into small, inspectable decisions.
- [Recipe 1: route support emails](https://jev101.com/markdown/en/support-routing.md): Produce a reviewable label without sending messages or issuing refunds.
- [Recipe 2: filter RAG evidence](https://jev101.com/markdown/en/rag-reranking.md): Retrieve candidates first; distinguish relevance from factual support.
- [Recipe 3: connect Jev to n8n](https://jev101.com/markdown/en/n8n-workflow.md): Start with a generic HTTP Request node and an explicit review branch.
- [Recipe 4: select fields and validate them](https://jev101.com/markdown/en/structured-extraction.md): Separate candidate selection from parsing, arithmetic and validation.
- [Understand probability, confidence and review](https://jev101.com/markdown/en/confidence-and-fallback.md): Valid output types do not establish correct decisions.
- [For Chinese users: evaluate your own language data](https://jev101.com/markdown/en/evaluate-chinese.md): Language support is not evidence of reliability on your domain.
- [Estimate cost and measure latency](https://jev101.com/markdown/en/cost-and-latency.md): Include request size, retries and the complete workflow.
- [Before launch: permissions, data and execution gates](https://jev101.com/markdown/en/production-checklist.md): Separate a model judgment from permission to act.
- [Troubleshooting errors and incorrect decisions](https://jev101.com/markdown/en/troubleshooting.md): Identify whether the problem is access, protocol, judgment or execution.
- [Glossary: decode the terminology](https://jev101.com/markdown/en/glossary.md): Keep models, interfaces, integrations and workflows distinct.
- [Versions, sources and editorial rules](https://jev101.com/markdown/en/versions-and-sources.md): Distinguish documentation review, offline tests and live validation.

## Data and discovery

- [Resource catalog](https://jev101.com/data/catalog.json): 168 entries; original IDs, sources and bilingual descriptions.
- [Use cases](https://jev101.com/data/scenarios.json): 19 directions with editorial flow illustrations.
- [Chinese directory](https://jev101.com/zh/tools/index.html)
- [English directory](https://jev101.com/en/tools/index.html)
- [Sitemap](https://jev101.com/sitemap.xml)

## Citation policy

Cite the specific stable guide or project URL. Preserve snapshot dates and distinguish source claims from this site’s editorial suggestions. No guarantee of indexing or citation by any AI service.