jev-crush
Web app that reads a two-person chat and labels each message's emotion, intent and reply quality to estimate interest.
Overview
jev-crush is a Chinese-language web app derived from Crush Monitor. You paste a two-person chat (WeChat, QQ or WhatsApp exports), mark which side is you, and Jev labels the other person's messages with emotion and intent, rates your replies, and combines them into an overall interest score with a suggested next step. Labels appear directly on the chat bubbles rather than as a written essay.
Relationship scenes (romantic, friends or colleagues, social platforms) change how scoring is interpreted. The README is candid that numbers are model judgements, not proof of feelings, and that the model sees only text, not tone or offline context. It uses React, Express and Zod, stores data as local JSON, supports TypeSafe, Vercel AI Gateway or OpenRouter as the Jev provider, and includes an admin area and optional redeem codes. Chats are sent to the chosen provider and synced records are visible to admins.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What emotion does this message express?(inferred) | — |
| Choice | What is the intent of this message?(inferred) | — |
| Score | How good is this reply?(inferred) | — |
How it works
- Paste a two-person chat and choose which side is you
- Parse speakers and timestamps
- Ask Jev for emotion and intent per other-side message and a rating per own reply
- Combine into a conversation-level interest score and suggestion
Strengths
- Per-message labels instead of a generated essay
- Scenario-aware scoring
- Choice of three Jev providers
Caveats
- Sends private chat text to a third-party provider
- Interest score is a model judgement, not ground truth
- Text only; images and voice ignored
Getting started
Clone, `npm ci`, copy .env.example to .env with JEV_API_KEY, then `npm run build && npm start` and open http://127.0.0.1:3178/.At a glance
- Type
- Project
- Status
- prototype
- Language
- TypeScript
- Runtime
- React and Express web app
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Jev-crush：把一段双人聊天读成情绪、意图和好感信号。贴进微信记录，标签标在气泡上。Jev 判断，MIT。
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Emotion and intent of each message from the other person, a rating for your replies, and inputs to an overall interest score.
Is the score reliable?
The README says it is the model's judgement, not proof of how the other person feels.
Which keys work?
A Jev key from TypeSafe, Vercel AI Gateway or OpenRouter.
Related projects
DeepSeek Harness plugin that classifies each user message's emotion, urgency and department with Jev and injects it as context.
PluginbetaTypeScriptZero-dependency Android app that judges the visible WeChat message's intent, emotion, urgency and reply posture with Jev.
AppbetaJavaAndroid WeChat assistant where Jev judges the chat and scores the relationship, and a chat model drafts three replies.
AppbetaKotlinWeChat plugin that classifies a long-pressed message's intent, emotion, urgency and suggested reply posture with Jev.
PluginbetaJavaLocal chat analyser that labels each WeChat message's emotion and intent with Jev and grades how your replies landed.
ProjectprototypeJavaScriptWeb tool that analyses Telegram channel JSON exports locally for intent, quality, sentiment and speaker tone with Jev.
Articleprototype