jev-哑巴微信
macOS WeChat helper where an LLM drafts replies and Jev classifies intent, rates risk and ranks the candidates; you press send.
Overview
jev-yaba-wechat is a macOS assistant that floats beside the WeChat window. It captures the window, recognises chat text, message sides and group senders locally with macOS Vision, and then splits work between two models: Jev judges and GPT writes. The judgment layer classifies the other person's intent into eight categories (assigning work, chasing progress, asking progress, criticism, asking for explanation, small talk, scheduling a meeting, praise) using up to the last two messages, shows a 0 to 9 communication-risk level with suggested response direction, and ranks the candidate replies.
The generation layer writes two replies per selected tone, from ten built-in styles plus custom ones, for up to six candidates. You can copy a reply or have it filled into WeChat's input box through Accessibility, with read-back confirmation; sending is always done by you. Without a TypeSafe key, or if the call fails, judgment falls back to a local decider-2b model of about 7 GB. It is source-only for macOS 13+ with documentation in Chinese.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What is the other person's intent? | assigning work · chasing progress · asking progress · criticism · asking for explanation · small talk · scheduling a meeting · praise |
| Score | How risky is this communication (0-9)? | — |
| Score | How suitable is each candidate reply?(inferred) | — |
How it works
- Capture WeChat window and OCR locally
- Jev classifies intent and rates risk
- GPT drafts replies in chosen tones
- Jev ranks candidates; user fills one in and sends manually
Strengths
- Clear split: Jev judges, LLM writes
- Never sends messages automatically
- Local OCR and local judgment fallback
Caveats
- Local fallback model download is about 7 GB
- Source-only, macOS 13+
- Model judgments of intent and risk can be wrong
Getting started
git clone https://github.com/wuxie888/jev-yaba-wechat.git and follow the setup in the README (macOS 13+).At a glance
- Type
- Project
- Status
- prototype
- Language
- Swift
- Runtime
- macOS app
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
This project integrates Jev to provide structured decisions for its workflow. See the repository for implementation details.
logicrw/awesome-jev-projectsmacOS helper beside the WeChat window: an LLM drafts several possible replies and Jev scores them, leaving you to press send. Chinese.
MrJev/awesome-jev
Frequently asked questions
Does it send messages for me?
No, you choose a reply and press send yourself.
What does Jev do versus GPT?
Jev classifies intent, rates risk and ranks candidates; GPT writes the replies.
What if I have no TypeSafe key?
Judgment falls back to a local decider-2b model.
Related projects
- jev-chat-windows★ 802
Windows WeChat reply assistant: local offline OCR reads the chat, Jev judges intent and tension and ranks three drafted replies to fill in.
ProjectproductionPython AstrBot plugin where Jev decides whether the bot should join a group conversation; the main model still writes the reply.
ProjectproductionPythonDeepSeek Harness plugin that classifies each user message's emotion, urgency and department with Jev and injects it as context.
PluginbetaTypeScriptiOS custom keyboard that shows a copied message's intent, risk level and candidate replies inside any chat app, filling but never sending.
AppbetaSwiftWindows desktop chat assistant that OCRs a selected chat region, has Jev analyse the conversation, and ranks drafted replies for copying.
ProjectbetaPython, TypeScriptmacOS chat overlay that reads the current chat window and shows the incoming message's intent, risk and candidate replies, read-only.
ProjectbetaPython