Jev by TypeSafe
Fast, cheap yes/no, classification and rating decisions for AI agents with Jev, TypeSafe's System One model: typed answers with calibrated probabilities in about 300 ms, instead of a paragraph to parse.
7 天免费 · 无需信用卡 · EU 托管
问 Jev 任何事
通过 AnythingMCP 对 Jev by TypeSafe 试一个真实提示 — Claude 调用工具并返回答案。在你的消息上按 Enter 即可在 Cloud 安装。
在 Cloud 上安装 JevClaude 是 AI,可能会出错。请核对回答。
2 分钟设置
- 1在 cloud.anythingmcp.com 开始免费试用。
- 2在连接器商店点击 Jev by TypeSafe。提示时粘贴凭据。
- 3生成 MCP API 密钥,将 AI 客户端指向生成的 MCP URL。
配置 Jev by TypeSafe
直接来自连接器定义:在哪里获取凭据、需要哪些权限,以及该适配器调用的是哪个 API。 由连接器作者以英文提供。
Jev is a decision model, not a chat model. You give it state (the evidence) and one or more typed questions; it returns a number or an option with probabilities in about 300 ms. It never writes text. Use it when an agent needs a judgment it can branch on: is this urgent, which team owns it, how risky is it, does this reply answer the question, is this record a duplicate.
Start with jev_playbook. It costs no API call and explains how to phrase questions, what to put in state and how to read the probabilities.
Setup: create an API key at https://console.typesafe.ai and paste it into TYPESAFE_API_KEY. Jev is billed per input token (about $0.04 per million), output is free.
Tools
jev_yes_no: one yes/no question, returns the probability that the answer is yes.jev_classify: pick one option from a set you define, with a probability per option and a confidence.jev_rate: place the state on ordered levels (2 to 10), returns a 0-indexed score, the level probabilities and a legend.jev_ask: the full API: several questions of any type over the same state in one call (they run in parallel and cost the same state tokens once).jev_list_models: model names your key can use.
Question types (the API calls a yes/no question noul, never bool): noul (probability 0 to 1), choice (one option, max 255), score (ordered levels, max 10).
Reading answers: a yes/no value near 0.5 means unsure, not medium. confidence on choice and score measures how concentrated the probabilities are, not whether the answer is right. A score over N levels runs from 0 to N-1: 2.6 on 4 levels sits between level 2 and 3. Decide thresholds in your own logic: act on confident answers, re-ask or escalate to the user when it is unsure.
Limits: text only (no images), 32k tokens for state plus the longest question, 64k per request. English is the strongest language; German, Italian and others work but check confidence.