Jev 创始人推荐的 10 个决策模型 GitHub 仓库

morph @morpphhhaw原推发布于 2026.10.01

在 X 上打开原推

this is a f*cking treasure for JEV devs and I'm not kidding Jevs founder Diogo Almeida just mentioned his favorite github repos people are putting small decision models inside coding agents, QA systems, desktop control, smart homes, YouTube filters and real-time games these 10 make the architecture click: JevAny - open infrastructure for training, evaluating and deploying your own System One decision models ↳ http://github.com/SimpleJev/JevAny jev-stars - turns your GitHub stars into searchable memory for coding agents, with cached Jev tagging and a built-in MCP server ↳ http://github.com/AkashPriyadarshii/jev-stars agentic-sop - runs an entire software-development lifecycle around coding agents, then uses Jev as a quality gate before human approval ↳ http://github.com/imhttran/agentic-sop jevqa - tests web apps from plain-English checklists; Playwright controls the browser while Jev chooses elements and judges each step ↳ http://github.com/ado11231/jevqa ha-gutcheck - performs weekly Home Assistant health checks and uses Jev to classify broken integrations, risky updates and missing device metadata ↳ http://github.com/funkadelic/ha-gutcheck claude-jev - a Claude Code plugin that sends prompt routing, rule checks and context compaction to Jev instead of a frontier model ↳ http://github.com/0x7067/claude-jev harness-router - places Jev directly inside the tool-selection loop for Codex and Claude Code, with confidence fallback and optional MCTS ↳ http://github.com/Protocol-Lattice/harness-router jev-touhou - a local Jev-pattern engine using Laya and Mica to play an open-source Touhou game one decision at a time ↳ http://github.com/NatBrian/jev-touhou FeedKeeper - a Chrome extension that asks Jev whether each YouTube recommendation belongs to a category you want removed ↳ http://github.com/Skynet843/feedkeeper jev-bot - lets Codex and ChatGPT control native macOS applications while Jev selects the right control from the accessibility tree ↳ http://github.com/stoopid-computers/jev-bot pick by what you are building: > coding agents → claude-jev, agentic-sop > tool routing → harness-router > testing → jevqa > local decision models → JevAny, jev-touhou > personal infrastructure → jev-stars, FeedKeeper > smart homes → ha-gutcheck > computer use → jev-bot the interesting part is not that all ten “use AI.” each project gives the model a closed set of answers, asks for one narrow judgment and leaves execution, permissions and state inside ordinary code that is the Jev pattern find one expensive decision your agent repeats hundreds of times and replace it before your next frontier-model bill arrives

这条是引用着另一条写的：x.com/n01ennn/status/2103542021071978601