拿到 Jev API 后可以照抄的 9 个项目

Brjan | AI Builder @MrBrjan原推发布于 2026.09.22

在 X 上打开原推

Jev is blowing up. If you just got API access and don’t know what to build, copy this list. 1. jev-ultrafast A fast browser agent from Browser Use. Jev only picks the next action and which element to click. A small model writes text when needed. One Google Flights search takes about 7 seconds. https://github.com/browser-use/jev-ultrafast 2. fast-jev-compaction Context compaction for Claude Code. After each tool call, Jev decides what is still useful. Junk gets dropped. Kept text stays word-for-word. https://github.com/tamaratran/fast-jev-compaction 3. json-render Vercel Labs’ generative UI framework. In the experiment, Jev does not write JSON token by token. It only picks components, props, and layout. https://github.com/vercel-labs/json-render 4. typesafe-mcp Best starting point if you just got the API. Plug Jev into Claude Code, Claude Desktop, Codex, and Pi so you can call Choice / Score / Noul anytime. https://github.com/itsmostafa/typesafe-mcp 5. jev-mcp Ready-made judgment toolbox for agents: fact checks, content screening, semantic ranking, classification, and extraction. https://github.com/jkudish/jev-mcp 6. SemDecide Turns Jev into a CLI. Classify, score, and filter from your shell. Good for crawlers, CI, and data pipelines. https://github.com/sharziki/semdecide 7. jev-codex-router Jev first scores how hard the coding turn is, then picks the model tier, reasoning depth, and speed mode. https://github.com/0xNatoshi/jev-codex-router 8. Winnow Context garbage collection for Claude Code. When Read / Bash / Grep dump a huge result, Jev decides what actually matters for the current task. https://github.com/GhalebDweikat/winnow 9. jev-review Run the diff through Jev first. High-risk changes get flagged, then a bigger model or a human reviews them. Comes with a local dashboard. https://github.com/devagrawal09/jev-review 10. Blink Use Jev as a codebase navigator. At each folder, it picks the files most relevant to the question, then walks deeper. https://github.com/ellipsis-dev/blink 11. agent-desktop Desktop automation. It reads the system accessibility tree, then Jev picks the next button, menu, or input field. https://github.com/lahfir/agent-desktop 12. typesafe-mario Jev plays Super Mario Bros. No screenshots. It reads structured emulator RAM, then decides whether to run, jump, or dodge. https://github.com/fhshaik/typesafe-mario 13. jev-drone Jev flies a drone at the high level. The flight controller still handles stability and safety. Jev only decides things like climb, brake, or go around an obstacle. https://github.com/RomanSlack/jev-drone 14. OneVOneJev A 1v1 FPS in the browser. Every tick, Jev chooses movement, camera, aim, fire, and jump. https://github.com/emrickgarrett/OneVOneJev 15. jev-trader High-frequency market making on the Monad testnet. Jev looks at the spread and trade direction, then decides buy or sell. Model latency is about 81ms. https://github.com/jarrodwatts/jev-trader 16. Prism Jev does not place the order. It scores states like toxic flow, market pressure, and mean reversion, then hands that off to your existing strategy. 17. neo4jev Drop Jev into a knowledge graph. At each node, it picks the next edge worth following, then keeps walking. https://github.com/jexp/neo4jev 18. jev-curate Filter training data with Jev. JSONL / Parquet rows get scored for quality, relevance, and risk before they enter the training set. https://github.com/AkashPriyadarshii/jev-curate 19. Canny Stops coding agents from claiming they are done. It looks at tool output, the code diff, and test results, then judges whether that “done” claim is real. https://github.com/qkal/Canny 20. killmyidea Feed it a startup idea. Jev scores it on several axes and returns KILL, FIX, or SHIP. https://github.com/monteduro/killmyidea

这条是引用着另一条写的：「Jev API 上手清单：15 个开源项目可直接抄」