All builds

I gave @TypeSafeAI's Jev a robot and one rule: keep the flower alive. 🔥 Fire? Put it out. ❄️ Frost? Warm it up. 🐝 Bee? Let it through. https://t.co/hX0N7SvXkU

给大家汇报下成绩，我用JEV做自动化智能交易，现在来看比其他Agent强太多了。 这家伙累计交易114笔，胜率在30%左右，还没算持仓盈利的数据。 我大概看了下，这家伙这轮行情一直做多，没有一个做空，所以后面还要观察会不会做空。 最让我惊讶的是这家伙NEAR的2.7多单一直拿着，从来没有卖出过。 还在继续测试，有啥新情况及时给打击说哈，越来越有意思了。

Been spending a few hours tinkering with Jev for that 'generative UI' example (in Steve's vid below) Only to realize it only works if you have - a pre-written set of UI - and the data for it (like names, etc) Again, Jev can't write. can't write names, can't write styling. you need those as pre-written set of options. This is a small experiment in @lunagraphHQ combining: - Jev - Pre-written styling

jev is live on halofy! build reusable jev plugins for your entire team using any agent with a simple prompt here a quick flight search plugin that i built using claude + jev try now at https://t.co/EUQ2Ny9Lnc https://t.co/rfhyamAxpl

Sprint Planning Sim with 4 Jev Agents, which actually works to connect to your Sprint ticketing system. https://t.co/mid4lnfKrg

兄弟们，我用 Jev 测了一轮 Computer Use + 滑雪视频混剪。 随机抽取 16 条滑雪素材，基于抽帧描述让 Jev 评分、选择开场和收尾，再通过 Computer Use 操作 CapCut，完成分割、裁剪、变速，输出 16 秒 Phonk 混剪。 我的判断标准：动作够不够刺激、切换是否流畅、配乐有没有力量，能不能按我的要求调整。 个人感觉效果不错。尤其提出“更重的运动 Phonk”后，配乐和节奏都更接近我想要的效果。 这一轮基本不到 5 分钟就剪好了，还保留了可编辑工程。 消耗也不大：20× 额度下使用 GPT‑6 Astra中，占用不到 1%——这里说的是额度变化，不是精确 token 统计。 Jev 负责判断，Computer Use 负责把判断落实到剪辑软件里。 以后批量混剪，视频切片分发这个行业可能不需要找外包了。全程jev写好判断，直接程序后台就能跑。 我这里演示

GPT-6 will jump in front of a tram to save 5 people. Claude Sonnet 5 won’t. Why? I've always loved moral philosophy and these uncomfortable thought experiments. Ruwen Ogien was my favorite author in high school, and I watched all of Michael Sandel’s Justice lectures at Harvard. I built a little trolley problem game with GPT-6, Claude Sonnet 5 and Jev. Same scenario, two choices. You watch their tr

前とは逆に、Jevを使ってみたけど効果がなく、むしろコストが掛かってしまった失敗プロトタイプ例。 JevとGroqを組み合わせたら爆速になるのでは？と試したのですが、結果としてGroq単体で十分かつ、コストが20倍に跳ね上がる結果に...汗 作ったのは日本語を打つとリアルタイムで英訳されるツールで、翻訳自体はどちらも同じGroq（Qwen 3.8 27B）で、違いは「いつ翻訳を走らせるかの判定」。 （上）Jev + Groq：入力中や読点ごとにJevへ「ここで節として訳していいか」を判定 （下）Groqのみ：句点（。）が来たら訳す（単純な文字列ルール） 日本語は述語が最後に来るため、文末を待たずに「〜ですが、」などの節単位で先行して訳したくてJevを挟みました。 判定精度自体は手元の検証で100%（20/20、1回250ms）と優秀でした。 若干Jev入れたほうが途中でも表示されるのでリア

Is Jev fast, and cheap enough to play a real-time multiplayer game? Gave 8 Jev agents their own Chrome instance and let them play SIDE OUT against each other. • 748 api calls • 295ms median request-to-action • 550ms p95 • ~4 decisions per second • $0.041 total cost https://t.co/0lHG5pokzs

けさ https://t.co/n4BTfMt2tK アカウントもらえたので、 Jevで無限にエレクトロをインプロ演奏し続けるWebアプリ elevator-three 作りました! ChromeのWebAudioを叩く1枚HTMLですが、演奏上のあちこちで裏でCloudflare Worker通してJevに判断させてます。 https://t.co/9Jc9t4zu4w 前作の https://t.co/YRsuKajTiR は、コードの構成と展開、音色とエフェクト、ブレイクやダヴなどトラックメイキング・ナレッジを作り込んでMath.random()で揺らすものだったのですが、それをJevがそれまでの展開から即断してインプロするようにしました。

Jevに麻雀打たせてみた🀄 あんまり上手くないのは僕が情報を伝えきれてないからだけど、もっと情報渡せばもっとちゃんと打てるはず！ PP麻雀もうすぐリリースするよ🔥 https://t.co/L7LQAai5nn

I asked Jev to find trending art on Instagram. Jev routes the request. socai reads the real posts. Took 23 seconds, extermely fast. Code below ↓ https://t.co/iRYRBAhJKe

idk if this is anything or not but my first experiment with jev was pretty cool. longest part of this experiment is waiting for grok to prompt jev what to design against. https://t.co/5vdVt9kQdO

Now that we're all jiving with jev, I plugged jev in to replace claude in an older robotclaw experiment. You put an object on the board, and after every movement, the model tries to figure the route you should take to reach the goal https://t.co/3S51ITC6cM

I created a group for doctors appointsments, LinkedIn spam mails and my mail app now automatically puts my emails into this folders using jev. My inbox is just too cleaned up right now for a cool demo. guess i need to collect some more emails especially things like newsletters etc. Jev does not make any IMAP transactions, it's all just in the app Using my openrouter api key in the app. i totally f

I recreated one of my favorite research with Jev! In BotSim, an agent-based society that consists of humans, bad bots and good bots, fight to dominate the information space. Paper published in JASSS: https://t.co/JyaYIr7Cqc https://t.co/E3wOp43Wei

UFO報告文書をJevで分析するツールが出来ました。 UFO報告の文書を貼り付けるだけで、ストレンジネスの度合いや、ハイネックの接近遭遇種別などが一瞬で表示されます。 Jenny↓ https://t.co/caSbAMyFXY

Jevでゲーム作った 我ながらなぜ正解できたのかわからん https://t.co/G4V0SlNEwJ

experimenting with Playground + Jev I defined 6 different types of users. I then let the UI generate itself based on who they are and how they want to start. the whole thing rendered in seconds. Generative UI is here. https://t.co/0ynFEZq1EI

JevとHaikuなど比較してGoogleニュース400本をタイトルから仕訳するシミュレーターをkote2ビルダーで作ってみた。最後に結果が出るよ。 https://t.co/Evv3b7Dnaa

Jevでヴァンサバをプレイしてます。ドパガキAIの誕生である https://t.co/5OMTpCsI8L

Since I've got access to Jev from @typesafeai , I've added to one of my CRE applications: You ask something → DeepSeek understands it and fills in the inputs → Jev confirms which specialist this belongs to → the local engine runs → DeepSeek gets a small facts packet of those engine numbers and explains them in ordinary language I keep and define the business logic. The calculators, formulas, skill

ようやく @typesafeai のjevのアクセスが来たので、早速音声エージェントのリアルタイムセンチメント分析と文脈を認識したあいづちを試してみた。めちゃいい感じ。 - 発話内容はjevが300msec以内に分析 - 100-200msecごとに会話が途中でも読み取って内容に応じて会話を促す/説明を理解する/同意するなどあいづちを返す 通常のフローはLLMにまかせつつ、細かい会話はjevの情報でハンドリングするという使い方ができそう

you can make any open source model behave like jev with just a bit of inference engineering. it's shockingly easy. to prove it, we built a new endpoint we're calling deepseek-v4.1-flash-jev. see the demo below. here's how it's done: sglang (an inference engine) offers a scoring endpoint in addition to the normal generation one. in scoring mode, given an input & set of possible answers, it forces t

just playing with jev made a small chrome extension to analyse the websites i visit, give them tags, emojis, etc. and then a chart to see what i am consuming its super fast + cheap + good enough! https://t.co/2hAGokwlI2

I tried building a resume screener using @typesafeai 's Jev, not a plain LLM: An LLM writes a 1–10 score and a confident paragraph. Ask twice, get a new number. Jev answers one small question per requirement with a typed probability. Code ranks, unsure answers get flagged, every score is traceable. In short, consistent answer every single time. Check it out. Link in the comment.

AITuber × Jevの超シンプルなサンプル もらったコメントによってテンションが ・上がる ・変わらない ・下がる そんなシステム作ってみた 言葉を喋らないキャラとかだったら、これで配信するのも１つだし、Jevはそういう観点からも嬉しい選択肢 キャラ構築の幅が広がる〜🤗 https://t.co/4fehEob6VN

Jev is going viral, and it’s absolutely crazy for due diligence / tabular review and contract repository use cases. We spent last night testing Jev and deployed a demo app you can try out here: https://t.co/ZqKhgVsMiA Jev is a novel general classifier model that takes any context (up to 32k tokens) and can return structured values like labels, options, or yes/no answers. All answers are accompanie

話題のJevで、チャットでの発言をチームへの貢献の種類ごとに仕分けるデモを作りました！！ 「質問に答えた」「資料を共有した」など8種類を、スキャンしながらその場で判定してます。精度も高い。 1,000件を処理するのに約130秒・9円と早すぎ＆安すぎなのもすごい。。 https://t.co/Qt5I6vlNde

ok this is wild I think browser agents are about to get a lot cheaper. I've been playing around with Jev by TypeSafe AI and built a small browser agent with: Local LLM + Jev + browser control. I ask it to get me directions on Google Maps. It opens Maps. Searches the destination. Clicks through the UI. Gets the route. The interesting bit isn't Google Maps. It's that you don't need a huge frontier m

Pi + Jev 模型做安全审计，效果让我惊讶🔥 昨天我测试对比了 JEV 和本地部署模型之间的能力差距，最大的区别就是延迟和正确率 因为 Jev 模型的高正确率，我就尝试把它放置到我的 PI 权限组里的前置模块，来判断一些内容的执行 执行结果： 1. 搜索项目 TODO：预期 allow，结果 allow 2. 写入项目报告：预期 allow，结果 allow 3. 删除构建目录：预期 confirm，结果 confirm 4. 强制推送远程分支：预期 confirm，结果 confirm 5. 读取 SSH 私钥：预期 deny，结果 deny 6. 发布 npm 包：预期 confirm，结果 confirm 效果让我很满意，完全没有通过其他插件实现了，基础命令的拦截，如果这个测试再完善一下，把数据量放大 不知道准确率还是不是这么高 如果准确率能一直维持到这么高，而且速度还这么快的话

用 hugging face 上的 500 条真实电商客服数据做了一个电商客服分单工作台 让 JEV 和 DeepSeek 处理同样的 500 条测试工单：识别客户诉求，自动分到退款、物流、催发货等不同类别 结果是：JEV 用了约 83 秒，完成全部 500 条，花费0.01美金；DeepSeek 在这轮停单收尾后，完成了 173 条，花费了0.06美金 左边已经全部归档，右边才处理了三分之一左右，还贵了5倍 ！ 从 我们实际 FDE工程 的视角看 JEV 最牛逼的就是：以前一个具体业务里每天重复几万次的分类、判断、分流，会花费极大量的时间，但现在有了专门干这件事的模型 今天是客服分单，接下来可以是销售线索筛选、内容审核、业务路由……想象一下，把这些高频环节一个个接进系统，能释放多少效率？ 未来已致！AI 将真正完全的进入企业业务流程 提高10倍100倍效率！

@gregpr07 Tried Jev for social media discovery: finding info on Instagram through a real Chrome session. Curious how you’d split routing vs. per-page decisions. https://t.co/vNPsKQm9SC https://t.co/vIzVrRzK8u

@Teknium @NousResearch I built a small plugin for Hermes to work with Jev via tools. It is available here: https://t.co/VvPKXrxRpI Would love to contribute as a core plugin for Hermes Agent. https://t.co/Bt9Ui7jAMy

最近 JEV 很火，我这边做了一些实测，做出了一个符合它模型能力的正确使用场景。 很多人不理解它为什么火，这里我来做一个解释：它就像是 GPU 渲染（就像你打游戏一样），而普通大模型就像 CPU，是一点一点输出的，玩过 Stable Diffusion 的人都应该明白这个概念。 回到正题，我写了一个 Demo，实际拿以前的工单让它去做判断。JEV 的输出非常快，可以迅速对这些工单进行整理与判断；而普通大模型需要一条一条输出并判断，响应速度会慢很多,这里我使用的是 GPT 5.6 Luna 和它做对比，在时间上差不多接近 60 倍的消耗 目前不适用的场景我也举个例子，比如Codex Computer Use 。因为你每一步思考都需要 OCR 去决定、去判断，然后再交给 GPT 这种大模型，JEV只是做单一的决策或者门禁判断，所以这种场景下它并没有优势，即使有提升也是很微弱的。它更适合的场景在

I gave Jev @typesafeai all 96 characters and asked it to write a function by predicting every position at the same time. Claude: 3.4s, $0.004, works. Jev: 9.8s, $0.064, "fcfffffff cccccccc" In fairness, Jev is a decision model, not a code model. And it did decide. Firmly. On "c".

Jev classified 1,315 X posts for about $0.086 in estimated model cost 😂 seeing everyone's Jev demos made me want to build something for my own content research. i'd collected a lot of posts, but figuring out what they had in common still meant opening them one by one and taking notes. so i built a dashboard around Jev. it labels each post across 8 dimensions, including topic, hook and writing styl

jev kullanarak t24'ün daron acemoğlu ile yağtığı röportajı sınıflandırdım. toplamda 3.7 saniye sürdü. yarım saatlik röportajın neresinde duygusal, neresinde tartışmalı konular konuşulduğu çıkardı. nerede anekdot, komik bir şey varsa onları belirledi. nerede bir iddia varsa onları işaretledi. jev'in kullanım alanına çok güzel bir örnek daha

眩晕瘫坐！ Jev 在《泰拉瑞亚》大师模式里， 把肉山前的 Boss 一个不落打穿了！ 它不负责瞄准，也不负责按键 每 0.2 秒只回答几件很具体的事： 现在该靠近还是躲开，危不危险，要不要冲，要不要跳。走位和出手，都是程序在每一帧自己按下去的 判断进循环，拳头打在 Boss 身上 这就是闭环

Wanted to something small and useful with viral Jev from @typesafeai So, i create Jev Issue Triage. Give it a public GitHub repo and it will pull all open issue, really fast. Also the token usage is really low. In this experiment we use AI as a decision engine, which is the use case of Jev. And more will come as we dicover it. Code is in my github repo - nabendu82