Overview & result
Crush Monitor is a small tool that uses Jev to analyse a WeChat conversation with someone you like. According to the README it labels each message with its most likely emotions and intents, shows an overall interest score, grades your own replies and suggests a next step. It runs on your own machine with your own API key, and the author says the results are for fun and reference only.
Show the original post (Chinese)
Crush 好感监控器：用 Jev 分析微信聊天的情绪、意图和回复表现。本机部署，使用自己的 API Key。
How Jev fits in the loop
- Paste a chat log from WeChat, QQ or WhatsApp into the local web page.
- Choose which nickname is you and start the analysis.
- Jev labels each message, showing the three most likely of 12 emotion and 35 intent categories.
- The page shows an interest score and grades your own replies from SSS to D; pasting new messages updates the result.
How to reproduce
- Inspect the original showcase and source material at https://github.com/FerryCorleone/crush-monitor
- Verify the bounded prompt and input candidate schema configured for Jev
- Benchmark decision latency and classification accuracy against baseline models
Why this build matters
The author's reason for choosing Jev is that the tool needs a judgment per sentence, not long generated answers, and emotion and intent can be asked in the same request.
Limitations
- The author says the AI knows nothing about the relationship outside the chat and the result is for fun and reference.
- You need your own key from TypeSafe, Vercel AI Gateway or OpenRouter; Jev is a paid model.
- Only tidied text formats are supported; native iMessage bulk copy is not verified.