盘点 10 个用 Jev 做 agent 决策的开源项目

Arindam Majumder 𝕏 @Arindam_1729原推发布于 2026.10.04

在 X 上打开原推

Jev is getting a lot of attention right now. But after looking through the projects people are actually building with it, I think the more interesting story is not Jev itself. It’s the role it can play inside an agent. A few examples stood out to me: 1. browser-use/jev-ultrafast (~16.6k⭐) Jev picks the next browser action without needing another expensive LLM call. https://github.com/browser-use/jev-ultrafast 2. tamaratran/fast-jev-compaction (~6.5k⭐) Uses Jev to decide which context is still useful and what can be dropped. https://github.com/tamaratran/fast-jev-compaction 3. TheoLeeCJ/SemIf (~4k⭐) An open-source, local implementation of Jev-style decision making. https://github.com/TheoLeeCJ/SemIf-OpenJev 4. jarrodwatts/jev-trader (~2.1k⭐) A real-time trading loop where Jev chooses between buy, sell, or wait. https://github.com/jarrodwatts/jev-trader 5. kerpopule/hermes-jev-skills (~670⭐) Uses Jev for routing, memory, compaction, and skill selection. https://github.com/kerpopule/hermes-jev-skills 6. devagrawal09/jev-review (~570⭐) Code review where Jev helps make decisions instead of relying on one massive prompt. https://github.com/devagrawal09/jev-review 7. thruwire/foreman (~520⭐) Uses Jev as a supervisor for coding agents: continue, retry, verify, or stop. https://github.com/thruwire/foreman 8. superagents-lab/jev-search (~400⭐) Jev decides where an agent should search and how it should search. https://github.com/superagents-lab/jev-search 9. droidrun/mobile-jev (~350⭐) Uses Jev to choose the next action for Android agents. https://github.com/droidrun/mobile-jev 10. gargpratyush/jev-router (~340⭐) Uses Jev to route different agent tasks to the right model. https://github.com/gargpratyush/jev-router The pattern across these projects is pretty interesting: 1. LLMs generate. 2. Agents execute. 3. Jev decides. And that decision layer can show up in very different places: → Which action should the agent take? → Which context should it keep? → Which skill should it use? → Which model should handle the task? → Should it retry, verify, or stop? → Where should it search next? This is probably the most useful way to think about Jev. Not as another model that replaces your agent stack, but as a decision layer that sits inside it. I collected these 10 repos as a starting point if you're trying to figure out what you can actually build with Jev. Which one would you add to this list? https://x.com/0x_rody/status/2106413848982880695/video/1

这条是引用着另一条写的：x.com/arindam_1729/status/2103468454083002537