jev-wingman
Android chat co-pilot that reads the active chat via Accessibility and returns typed verdicts on intent, emotion and risk; never replies.
Overview
Jev Wingman is an Android app that helps you read a chat situation before replying yourself. Through Android Accessibility it reads the current chat window and, when you tap a floating button, returns structured judgments in a card beside the chat: whether the other person is waiting for a reply, their intent, emotional intensity, the risk of saying the wrong thing, and a suggested strategy. It does not generate replies, inject text or send messages, does not log in to or take over any chat app, and does not keep full chat history.
It offers two modes: remote Jev, which sends the last few lines of conversation to the endpoint you configure, and an offline local mode built on a quantised bge-small-zh-v1.5 sentence-embedding approximation, which the README states is not Jev's weights and does not claim official accuracy. QQ private and group chats are verified on real devices; apps that expose text to Accessibility usually work, while WeChat's custom-drawn chat view usually cannot be read. Version 1.9.1 adds a new console, permission fixes and demos. It is distributed as an APK and documented in Chinese with an English summary.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is the other person waiting for a reply? | — |
| Choice | What is the other person's intent? | — |
| Score | How intense is their emotion / how risky is it to say the wrong thing? | — |
How it works
- Read visible chat text through Accessibility
- Send recent lines to remote Jev or run the local model
- Show typed verdicts and a strategy in a floating card
- User writes and sends their own reply
Strengths
- Never sends or writes messages
- Offline local mode available
- Results bound to the captured conversation
Caveats
- Local mode is an approximation, not Jev
- WeChat chat view usually unreadable
- Remote mode sends recent chat lines off-device
Getting started
Download jev-assist-v1.9.1.apk from Releases, enable chat reading and the floating button, choose remote Jev or local model.At a glance
- Type
- App
- Status
- beta
- Language
- Kotlin
- Runtime
- Android app
- Has code
- yes
Use cases and integrations
How curated lists describe it
基于 Jev 的聊天决策辅助，不挑 App（QQ / 微信 / 飞书皆可）· An on-device chat co-pilot that returns typed verdicts instead of prose, built on Jev.
hellogumbo/awesome-jev
Frequently asked questions
Does it reply for me?
No. It only gives judgments; you write and send replies yourself.
Does it work offline?
Yes, a local mode uses a small embedding model, which the README notes is not Jev.
Which apps work?
QQ is verified; apps exposing text to Accessibility usually work; WeChat's chat view usually does not.
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 Self-hosted helper for tricky chats and everyday decisions: Jev makes typed checks, an LLM handles language, the user decides.
ProjectprototypeTypeScriptAndroid chat copilot: Jev judges intent, needs and risk in visible chats, then ranks drafted replies to fill, never send.
HookDeepSeek 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.
AppbetaKotlin