User repository on GitHub
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。
- Forks
- 27
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Tools & CLIs
Crush Monitor
A local React and Express web app that takes a pasted two-person chat transcript and annotates each message with emotion and intent probabilities plus a grade for your own replies. Chinese interface, your own API key.
- Stars
- 288
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- FerryCorleone
- Last verified
- Oct 2026
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Per-message labels show the top three of twelve emotions and top three of thirty-five intents, and your own replies get an SSS-to-D grade with a next step.
- Six weighted components produce the favorability score, an explicit refusal caps the total, and the README says it is not a probability of being liked.
- Runs locally on Node 22.12 or newer through npm ci, npm run build and npm start, serving 127.0.0.1:3178, with the provider and key set in the page.
- Handles WeChat desktop, QQ and WhatsApp export text; anything else must be reshaped to 'Name: line', and it never guesses the speaker or reads iMessage data.
Watch out
Analysis is Chinese-only, so the English README does not mean the interface was translated, and the README states there is still no cross-model accuracy evaluation for Chinese dating chat, which leaves the scores without measured reliability. Chat text goes to whichever provider you configure, so local storage is not offline analysis, and the key sits as owner-only plaintext in .runtime/model.json rather than in a vault. Each request is capped at 500 messages and 12,000 characters, one paste over 250,000 characters must be split, and images, voice notes, ZIPs, HTML and chat databases are not parsed. 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.
The README recommends Jev first and keeps DeepSeek or Qwen as a fallback with the same features. It warns that TypeSafe no longer grants new users free credit, that Vercel's free monthly tier currently cannot be used for Jev, and that OpenRouter treats Jev as a paid model with no free balance, so confirm your account balance and model access in the console before you start. Chat subscriptions do not count as API credit.
The key is written to .runtime/model.json in the service directory as owner-readable plaintext, never in browser storage, and the settings endpoint does not return it in full; the README is explicit that this is a local config file rather than an encrypted vault. Chat and analysis records sit in your current browser's local database, are not shared across browsers or ports, and disappear when you clear browser data.
Only the documented shapes work: WeChat desktop's nickname, time and body triple, QQ's nickname-colon-timestamp, WhatsApp .txt exports in either common layout, or text you have already rewritten as 'Name: line'. The program will not infer who spoke when only bodies come through, does not read the iMessage database, and handles two-person text chats only. Formats have automated test coverage, which the README says is not the same as testing every client.
No, and the README says so directly. Jev results keep the raw probability and certainty, while a text model returns relative weights that the code normalizes into a distribution, and those percentages are model estimates rather than calibrated probabilities. It also isolates your own replies from everything that happens later in the chat, so later turns cannot retroactively grade an earlier message.
Node.js 22.12 or newer, then npm ci, npm run build and npm start in the project directory on macOS, Windows or Linux alike. The service tries to open http://127.0.0.1:3178/ itself; set CRUSH_OPEN_BROWSER=0 on a headless machine. Keep the terminal running and afterwards npm start is enough. Changing provider or key happens in the page under chat settings and takes effect immediately, and switching model keeps the chat but clears old analysis so results do not mix.
Each model request stays within 500 messages and 12,000 characters, and a single over-long message is saved with a prompt to split it. A paste over 250,000 characters has to be added in pieces, and total history is bounded by browser storage space. Long conversations are processed in batches rather than being capped at 500 messages overall: appending analyses only the new content and rechecks recent messages from the other person, while your own older reply grades are kept.
Six weighted components: proactive continuation, responsiveness and investment, care and consideration, self-disclosure, intimacy of expression, and actual follow-through, with the itemised breakdown available by clicking the top of the page. An explicit and still-valid refusal caps the total. The README is direct that the score is not the probability that someone likes you. Scoring uses recent original text plus relevant quotes from history; events such as invitations, care, refusals and withdrawals are indexed, but old scores are never reused as evidence for new ones.
More Jev resources like this
361GitHub stars
A VSCode extension that debugs by driving the real debugger: Jev routes each paused step as a typed decision, and a generation model is called only when a breakpoint location, an expression, or a patch is needed.
Jev Tools & CLIs#community#jevbest
197GitHub stars
A Chinese-first desktop workbench for local models whose JEV page runs typed noul, choice and score questions against a local engine or a cloud key, alongside a gateway, RAG knowledge base and shared memory.
Jev Tools & CLIs#community#jevbest
Community
kunpengtalk/OmniStudioTypeScript
kunpengtalk/OmniStudioTypeScript124GitHub stars
Prism is an autonomous liquidity agent for Solana concentrated liquidity pools, currently Meteora DLMM. Every ten minutes it recalls past outcomes, reads pool state, scores volume authenticity and decides HOLD, REBALANCE, EXIT or ENTER behind a risk gate.
Jev Tools & CLIs#community#jevbest
38GitHub stars
Research dashboard for Polymarket BTC five-minute Up/Down rounds that feeds market state to Jev class scores and paper-trades the edge.
Jev Tools & CLIs#community#jevbest#catalog-import
38GitHub stars
Python memory filter that judges every stored memory against the current request with one calibrated yes/no per memory in a single Jev request, keeping everything above a threshold.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
samdotmak/jev-recallTypeScript
samdotmak/jev-recallTypeScript30GitHub stars
Browser extension that filters your X timeline with a model you choose and pay for, using plain-language rules plus per-author and blocked-word controls.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
tshmieldev/sharpTypeScript
tshmieldev/sharpTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
The request shape barely changes across surfaces; the surrounding contract does. When to call the API directly and when a gateway, a Worker binding, or a router entry earns its extra hop.
Jev Tools & CLIs#gateway#vercel#cloudflare
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Jev labels a record that already exists. Code moves, flags, or holds it. How the document, mail, and transfer tools keep the action out of the model.
Jev Tools & CLIs#classification#email#security
Read article
From the Jev community
Posts from builders shipping with Jev right now.