NewJev WeChat Reply Assistant
For a message and optional recent context, Jev receives two questions in one request: intent Choice and direct-reply-risk Score.
Added to Jevfast
NewHow it uses Jev
For a message and optional recent context, Jev receives two questions in one request: intent Choice and direct-reply-risk Score. The chosen intent maps to a static set of response actions. When GPT has generated reply candidates, a second dynamic Choice can rank them; that chart is omitted because concrete candidate texts are runtime-dependent.
What Jev decides
Illustrative source-derived request: message intent and reply riskExample answers · not a recorded Jev response · Source ↗
Question 1 · intent
YOUR APP
INSTRUCTION
What is the real intent of this message?
STATE
The state is the supplied message, prefixed by optional context and a blank line when context exists.
JEV · CHOICE
- 派活
- 催进度
- 问进度
- 批评
- 要解释
- 闲聊
- 约会议
- 夸奖
App workflow
- Detect and read a chat message
When read-screen mode is enabled, the app uses OCR on the local WeChat window; a detected message can trigger analysis.
- Judge intent and risk
Jev asks the two pinned questions together and the application maps the intent to static response guidance.
- Generate candidate wording
GPT generates reply text when configured; Jev does not generate the message.
- Review before sending
The user selects a candidate, fills it into WeChat, checks it, and sends it.
Why it is interesting
The implementation separates semantic judgment from language generation: Jev labels intent and risk, while GPT writes candidate replies and Jev can select among them.