20 个 Jev 项目合集:让 agent 选择、路由、过滤下一步
清单列出 browser-use、Vercel、MCP 连接器、代码审查、桌面控制、游戏和无人机等 20 个用 Jev 做结构化决策的开源项目。
this is pure f*cking treasure 20 Jev projects for agents that choose, route, filter and check their next action browser work. context. UI. model routing. code review. graph search. data. real-time decisions 01 jev-ultrafast browser actions and targets selected from the current DOM ▸ https://github.com/browser-use/jev-ultrafast 02 fast-jev-compaction score old tool records and retain selected text verbatim ▸ https://github.com/tamaratran/fast-jev-compaction 03 json-render + Jev select predefined components and properties for a UI spec ▸ https://github.com/vercel-labs/json-render 04 System One Connector (typesafe-mcp) give agents typed Jev decisions through MCP ▸ https://github.com/itsmostafa/system-one-connector 05 jev-mcp add evidence checks, content screening and candidate ranking ▸ https://github.com/jkudish/jev-mcp 06 SemDecide classify, score and filter text or JSONL from the terminal ▸ https://github.com/sharziki/semdecide 07 jev-codex-router classify each turn before local rules select a model configuration ▸ https://github.com/0xNatoshi/jev-codex-router 08 Winnow filter tool-output blocks while keeping original text available ▸ https://github.com/GhalebDweikat/winnow 09 jev-review score code quality and rank areas that need review ▸ https://github.com/NiazMorshed2007/jev-review 10 Blink score files and directories to guide repository search ▸ https://github.com/ellipsis-dev/blink 11 agent-desktop + Jev choose desktop controls and actions from accessibility data ▸ https://github.com/lahfir/agent-desktop 12 typesafe-mario choose legal controller actions from structured game state ▸ https://github.com/fhshaik/typesafe-mario 13 jev-drone select tactical maneuvers in a MuJoCo drone simulation ▸ https://github.com/RomanSlack/jev-drone 14 OneVOneJev choose movement, aiming and firing in a browser FPS ▸ https://github.com/emrickgarrett/OneVOneJev 15 jev-trader evaluate order-book state in a market-making experiment ▸ https://github.com/jarrodwatts/jev-trader 16 Prism log advisory market judgments alongside deterministic rules ▸ https://github.com/irfndi/prism-liquidity-agent 17 neo4jev score outgoing graph relationships and continue beam search ▸ https://github.com/jexp/neo4jev 18 jev-curate score dataset records and apply local filtering thresholds ▸ https://github.com/AkashPriyadarshii/jev-curate 19 Canny combine semantic completion checks with an execution ledger ▸ https://github.com/qkal/Canny 20 killmyidea combine dimension scores into KILL, FIX or SHIP labels ▸ https://github.com/monteduro/killmyidea the loop: observe the state → ask a bounded question → get a choice, score or probability → apply your rules → execute the action → check the result → feed the new state back Choice picks an option Score evaluates a rubric Noul evaluates whether a statement holds a practical starting point: choose one repeated decision in your agent, define the available actions, log its answers and compare them with the actual outcomes code examples, including experimental demos. Jev API access and setup are required save this, then read the full guide for loop + jev engineering ↓
这条是引用着另一条写的:x.com/beamnxw/status/2108101310515925217