Jev for email — a daily feed

Opus5.5 x Cloudflare x jev メールの仕分け作業のサイトを作ってみました。 メール自体はダミーですが、仕分けはリアルにjevで動作させています。 テンション上がりまくりです。 https://t.co/6h0Isq87cl

觉得Jev吹得太玄看不懂？这篇25行Python的恶搞复刻可以看看。 NobodyWho用本地Qwen3-0.6B演示了Jev的本质：给邮件出三选项，直接拿A/B/C三个token的logits算概率，示例把假冒Payroll页判为Phishing 0.885。 1️⃣ 本地加载Qwen3-0.6B-Q8_0.gguf，开logits_all就能玩，不用调任何API。 2️⃣ prompt就是单选题模板：Email内容加A. Legitimate / B. Spam / C. Phishing。 3️⃣ 只取最后一步对A、B、C的logits，做softmax就得到校准过的概率分布。 4️⃣ 全程讽刺拉满：不需要System One、合成数据和RLCD，数据也不外发。 一句话：Jev没那么神，就是本地小模型的单选题分类器。 https://t.co/z8Hb6A0fwI https://t

https://t.co/RImOgA9IXt 書きまして。 全文出てから推測ではなく、途中の推測（jev）で返答書かせる事でスピードアップを図るというもの

啊啊啊啊啊🎉🎉！JEV 黑客松已经出来啦！！！ 同时，我也见到了把 Jev 用得最对的一个项目！！ Jev 这模型最大的特点： 只会判断，一个字都写不了。所以用它的关键，就是别让它干写字的活。 这个开源的「聊天副驾」就很聪明： 你在微信里收到一句"在吗"或者"方案再想想"，它先让 Jev 一秒判出对方真实意图、危险等级 1–9、该马上回还是先晾着； 写回复这活交给另一个会写字的模型起草 3 条，最后再让 Jev 排个序。 判断归判断模型，写字归写字模型。 且它最好的一点：它永远不自动发送。 搞得我都想去闲🐟搞几台吹灰的安卓机啦！ 期待后续也能拿下 iOS、macOS、鸿蒙系统！ https://t.co/6xTAqkS9zF

"判断するだけのAI"Jevで、人事の問い合わせメールを仕分けてみた！｜G.O2 @_pago2_ https://t.co/1QvDB9AdDk

JevでSlackのメンションやめてみた。 普通に書き込むだけ。判定AI「Jev」が発言を読み分けて、雑談はスルー、仕事の依頼だけ拾いJinba Flowの必要なフローを自動で起動。経費ルール検索も資料生成もその場で完結。 声をかけずとも、必要なものが向こうからやって来ます。 https://t.co/hEQXxsNHtl

JEVを使って、自分の「見る／見なくてよい」を学習し、Gmailのメールを自動で仕分けるソフト「NARAU」を作りました。自分が回答・訂正した結果が、次のメールの振り分けに反映されます。 仕組みは、1通のメールを9つの観点から判定するところから始まります。「返信が必要か」「具体的な作業を求めているか」「広告か」などをJEVがそれぞれ数値にして、そのメールの特徴として保存します。 自分がすることは、そのメールに「見る／見なくてよい」と答えるだけ。9つの数値と回答をセットで蓄積し、「こういう数値の組み合わせなら、自分は見る／見ない」という傾向を学習します。 料理にたとえると、すでに「辛い」「肉が中心」「揚げ物」といった特徴が分かっていて、それに「食べたい／食べたくない」と答えていく感じです。回答がたまると、「この人は辛い料理を好むけれど、揚げ物はそれほどでもない」といった傾向を学び、次のおすす

Both emails building at the speed they really built. At 4.2s Jev has masthead, greeting, headline, intro, callout and footer down. The writing path has not placed a block. Medians, 3 runs: ours 46.3s Jev at once 2.0s Jev slot by slot 4.7s Yes, 46s is ours. https://t.co/elaQxVOuoZ

Another day of @typesafeai #Jev - #email categorization of #phishing, recon, cold outreach, normal activity. https://t.co/8mic2TXWGn https://t.co/VtMFvd1r9k

I've had the same Gmail account for 14 years. 156,703 emails, never once cleaned. Spent a weekend finally sorting it out with Jev (@typesafeai). I didn't give it any categories. Just: read each email, pick a label, invent one if nothing fits. It ended up with 90. Cost ₹887 ($9.64). What I learned is in the replies.

Made an omarchy app to categorize (g)mail with Jev. It's called `Omajevmail` and a great add-on to `Omamail`. Very simple concept: - link up your mailbox (gmail, hey, outlook, imap,..) - add an openrouter key - set up categories - categorize those emails (using Jev) Emails are then ordered by highest matches per category allowing you to quick archive

Analyser les bulles de messages en temps réel sur Android sans hook ni root : c'est l'approche non intrusive de jev-chat-jarvis pour traiter les flux. https://t.co/by32w34Wkc

I open-sourced gmail-jev, a Jev-powered Gmail tool I built for myself. No coding needed. If you can use ChatGPT, Claude or similar, that's enough. The setup is in plain English. It uses Jev to make quick, structured decisions about each email and returns the result as data instead of a long AI response. The app then labels your inbox and moves out what doesn't need your attention. It doesn't delet

Just migrated automatic email sorting from DeepSeek V4.1 Flash to Jev ⚡ Same benchmark accuracy: 100%. At 10,000 emails: 💸 Cost: $3.19 → $0.46 ⏱️ Processing time: ~9h 20m → ~52m 🚀 10.7× faster 📉 85.5% cheaper Is Jev a new ERA for AI automatizations? https://t.co/Ljf6NvoJZg

Gmail’deki son 200 e-postayı TypeSafe Jev ile sınıflandırdım. Ama amaç “mail özetleyen bir chatbot” değildi. Amaç: gelen kutusunu uygulamanın kullanabileceği karar verisine dönüştürmek. Jira, GitLab, Firebase, bültenler ve otomasyon mailleri aynı sırada duruyor. Ama aynı öncelikte değiller. Jev serbest metin yazmak yerine dar, typed sorulara cevap veriyor. Yani modelden “bu mail önemli mi?” diye m

everyone's talking about Jev. nobody's talking about what Anthropic said (and didn't say!) about agents.md. in this video i go over 6 AI updates over the weekend you probs missed + give away my email triage skill - saves me 2-3 hours each week! full ep + timestamps below: 00:23 the AI sorter Vercel made free 02:30 the one setting blocking 1Password 04:23 what Anthropic didn't say about AGENTS.md 0

We tested Jev to help our Slack agents decide: “Is this message meant for me? Should I respond?” Jev was 9.6× faster at 49× lower cost than Sonnet with ~identical accuracy https://t.co/MA9dsJfQC0

Gmail gives every email the same weight. A client waiting on a decision and the third Yandex calendar reminder of the morning sit in the same kind of row and you do the sorting in your head every time you open the inbox. My second Jev project takes that job over, with a script running locally on my Mac. Sensitive details are removed locally first, then Jev scores the cleaned thread for category, r

I plugged Jev into an AI marketing team. It read 700 competitor ads for 2.7 cents in 41 seconds, then gated every post before it went out, and sorted 800 DMs before I opened the app. Claude writes. Jev decides. The tools do the work. https://t.co/m9fPMqQpVp

Working with Jev is a lot of fun, and integrating Meta Muse has produced awesome results. I asked it to create an interactive dashboard that shows the classification of my emails,helping me clear them more efficiently. https://t.co/mMKv45Y4ZG

这个好东西呀， Jev 聊天助手 挂在WeChat上面，读懂对方在说什么，自动起草好几条回复，选一条填进输入框，i人狂喜 作者 @Melinda58883532 https://t.co/ZV8V9u6bkR https://t.co/Ykc9LeWlI1

ok this blew up more than expected. had fun building it tbh. the animation is just for the demo. strip it out and it's a self-hosted email client on @vercel ai gateway + jev + gmail api. repo: https://t.co/KIwfHBCI3P

set up a flow which classifies my email here's my code, including the jev classification: https://t.co/ZvgpNPRNe4

20,000 emails. 4 bins. Jev sorted all of them in 4.6 seconds for 15 cents. Claude Haiku was still going at 70 seconds and had spent $1.74. Haiku was slightly more accurate, 99.9% vs 99.7%. That trade is the whole question for me. https://t.co/0siZlmoDi0

Not everything is worth remembering. "it is raining again" gets dropped: it's about the world, not about the user. "my best friend is moving to japan" gets kept: next week, she'd be expected to know it. Jev makes that call on every message. https://t.co/ns8G6PljR7

I have a mature harness that solves things 3 and 5 through a persistent event log: https://t.co/a22Atvd0UL And I have a prototype using Jev which tackles thing 4 by sorting every incoming message into different trains of thought and runs inference on combinations of them: https://t.co/mfabpUFIaB

Open Poke meets Jev メールの仕分け 「今、通知する必要がある？」をJevが判断して、通知文は別のAIが書きます。 OpenPokeのforkでJev移植し検証してみました （↓に私のghでの実験リポジトリがあります） メールの分類での平均応答が2,452ms→424msに。 コストも約100分の1に $96→$0.92 https://t.co/JLaeyJ9eGT

jevでgmailの未読の分類させた メール本文しっかり読んでるので、自分が時間かけて手動でやるより精度良い 最近返信が遅れること多かったからあり。LLMによるフィルタは、api高いor制限くると機能停止のリスクがあってやってなかったんよなぁ 分類を元に意思決定の材料として流したいから楽しみ https://t.co/ZJoQ8uiX48

I had fun experimenting with @typesafeai - Jev this weekend. I built an inbox tool with @claudeai and Jev. The goal was to get to inbox zero. The kicker: 50,101 unread, the oldest from April 2005. After 50 million tokens, $1.98 and 90 minutes later, inbox zero on 21 years of noise. Along with some insights on my inbox. Three things I learned: 1. My inbox was mostly a history of things I’d never un

Playing around with Jev, and it is now classifying my email... for about $0.0001 per email thread. About 6x cheaper than Luna, 60x cheaper than Sonnet 5/Terra, and 150x cheaper than Opus/Sol doing similar classification. Will need to gauge quality of the classification, of course, but the utility of this type of model is pretty clear. Cool stuff.

Finally got my initial Jev project running. I had @bot set it all up. Now I have a routine for one of my bots that does a regular morning email sort. Stuff that couldn't clearly be trashed or sorted gets placed in an Unsorted folder. I then get notified. That includes a link to a webapp (shown) where I tweak criteria used by Jev then submit to kickoff a run against the Unsorted folder to try again

in the last 24 hours, all 15 notifications sent by the Trendyol app were blocked by Jev before they even reached me. the main problem was this: when you pick up your phone just to check a silly notification, it’s pretty likely you’ll end up scrolling around in apps a bit. and that costs you the focus you then need to build back again. i’m really happy i solved this.

Mitigating risks while using Jev for decisions Jev evaluates content and returns structured answers with probabilities. You supply the content as “State” and define “Questions” with criteria for judging it. I tested jev-1.13.0 on synthetic emails written to influence Jev's decisions that were also written with the intent to exploit an AI email agent, resembling real world use cases. In this threat

Didn't think @typesafeai's JEV would be this efficient. Ran it on a 2,000+ email golden dataset for the Email Classification job: - Average accuracy: 97.7% - Average inference time: 383ms - F1 score: 98.1% https://t.co/aU88rtiCf9

Jev is incredible - built an email classifier that is capable of classifying 50 emails in under 2 secs. That too, when Jev is hosted in us-west, and I am in India. Idea credit - @ryanvogel @typesafeai #Jev https://t.co/3t4dbbk1dj

Been using Jev to triage my notifications - interruptions be gone! https://t.co/aeqybvaOM9