User repository on GitHub
弦外之音 —— 微信聊天里的关系进展助手：读屏（无障碍树 / 截屏视觉）→ Jev 判读 + 攻略度 → 聊天模型出 3 条候选回复，攻略度常驻挂在屏幕上。不改微信、不发消息、不注入点击。
- Forks
- 2
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
JevGuide — Nisaka520
WeChat relationship assistant that combines screen reading, one 8-question Jev judgment with a strategy score, per-contact memory, and three chat-model replies behind a resident score bar.
- Stars
- 36
- Language
- Kotlin
- License
- MIT
- Published by
- Community
- Author
- Nisaka520
- Last verified
- Oct 2026
Works forJev for Classification
Used inJev for Software & DevTools
Highlights
- Two read paths: free accessibility tree plus screenshot-to-text vision for builds that expose no message nodes.
- One Jev request asks intent, emotion, urgency, reply need, risk, action, posture, and strategy score.
- Chat model turns Jev verdict plus memory into three replies, defaulting to steady, advancing, and playful across seven styles.
- Resident bar keeps the latest score with delta coloring; tap opens verdict rows plus copyable draft cards.
Watch out
Newer WeChat builds expose no accessibility text, forcing slower vision reads, and score wording must stay fixed or history stops being comparable. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Bystander judges and shows 3 Toasts with no memory or drafting. Guide adds a resident strategy bar, three sendable drafts, and per-contact memory with summary, facts, score history, and recent turns.
Auto tries the free accessibility tree first and only pays for a vision pass when it reads empty. You can pin tree-only or vision-only, but newer WeChat often needs vision at roughly seconds per read.
Jev returns 0-9 floats mapped by round(score x 100 / 9) because the API rejects 11 levels. Bands are fixed in the prompt: cold near 0%, normal near 50%, strong near 70%, close near 90-100%.
In filesDir memory files keyed by SHA-1 with an index, holding rolling summary, up to 12 facts, score history, and recent turns. Either dependency failing still leaves a partial result, and uninstall or one tap clears it.
Jev's score question type rejects more: an eleven-level scale was refused with HTTP 400 "Too many score levels. Must have at most 10 levels." So it uses ten levels and converts with round(score x 100 / 9), meaning 6.48 reads as 72%.
Each contact gets a JSON file under filesDir/memory/ named by an SHA-1 of the WeChat title. Memory holds a rolling summary up to 600 characters, up to 12 key facts, the last 50 strategy scores and 40 recent turns, all on the phone only.
The two external dependencies degrade independently: if Jev fails there is no judgment, if the chat model fails you still get the judgment, and memory still updates either way, so neither failure wastes the whole analysis.
Other listings of this project
Forks and copies of this project in the directory. This entry is the canonical one.
10GitHub stars
Curated discovery index for Jev launch posts and demos, browsable by application with original video frames and links back to every X source post.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
2456868764/jevguide
2456868764/jevguideMore Jev resources like this
15GitHub stars
Android accessibility app that reads visible WeChat single-chat text and asks Jev for intent, emotion, urgency, and posture, showing exactly 3 Toasts without writing replies or changing WeChat.
Jev Demos & Experiments#community#awesomejev#catalog-import
CommunityNisaka520/JevBystanderKotlin
Nisaka520/JevBystanderKotlin7GitHub stars
Android noise gate that asks a model whether each notification or SMS is noise and suppresses only explicit flags. Uncertain paths fail open, verification codes never leave the device, and sender blacklists stay local.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
ufec/jev-block-android-adKotlin
ufec/jev-block-android-adKotlin1GitHub stars
Cross-platform assistant that ranks pending messages by urgency with one Jev call per item, shown in a draggable overlay that snaps to the screen edge. Android reads system notifications via NotificationListenerService; Windows screenshots the WeChat window and OCRs it locally. It judges, never replies.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
jeffyuysw/jev-xianhuiKotlin
jeffyuysw/jev-xianhuiKotlin46GitHub stars
Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.
Jev Demos & Experiments#community#jevbest#catalog-import
941GitHub stars
Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
TypeLLM/TypeLLMPython
TypeLLM/TypeLLMPython98GitHub stars
Self-hosted Docker service and web UI that packages Laya structured-decision inference behind a compatible System One endpoint.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
1Panel-dev/laya-serverPython
1Panel-dev/laya-serverPythonFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.