X posts

Just vibe coded Jev search. Search any similar item using this extension. Just add your API key and you're good. Host it locally.. Repo link in the replies. Give it a try https://t.co/XGXh3W2Nyn

Jev + GPT-6 Astra is the best AI agent system I’ve ever built. CHEAPER and FASTER than 96% of the systems other people are running. Prompt → GPT-6 Astra → Jev decision → Astra execution → Result Here’s the setup in 7 steps which take just 6 minutes: Step 1 → Create a TypeSafe API key. Keep it in secure storage, never in chat. Step 2 → Add TYPESAFE_API_KEY to the environment running your Astra agen

you can't chat with Jev? who decided that? i built Jeff, a chatbot you can talk to! Jeff is a sassy chatbot who won't do your work and hates being spammed. Jev picks hand-written replies. no generated text. go talk to Jeff and burn my Jev credits. it's open source too, so build on it if you want. (links in thread)

Ok I got Jev + SAM 3 working on a simulated mobile robot! SAM 3 + camera depth provide 3D observations, alongside joint state and contact feedback. Code samples numerical movement, grasp and placement poses; parallel Jev calls rank them, then a final Choice picks the action. Noul and Score add supporting judgments, while deterministic controllers handle IK and motor control. It’s almost through th

とりまjevでゲームデバッグと APIコスト削減やってみた。 ゲームデバッグの方はかなりいい✨️ もう簡単なゲームのテストは人間いらなそう APIコスト削減はAstraのコスト減らせる？ と思ってたけど、そもそも0→1がAstra 機能改修5.6系でしてるからダメでした笑 ビックデータ系の修正検索用だね。 https://t.co/E1cq1yuqec

This is crazy 🤯 I ran jev on v3code. It one-shotted My launch video by decisions and its sick. https://t.co/o7899GSXWg

Mac Studio M1. Left is local Laya, right is cloud Jev. Same Snake, same questions, 30 seconds. Local decides in milliseconds. Cloud waits on a ~300ms round-trip. The scoreboard is just latency made visible. https://t.co/wqfEcw5gZe

Everyone's sharing Jev experiments, here's mine! Paste your @supabase RLS policies and watch Jev race gpt and gemini on them. Same rubric, same decision rule, side by side: who flags the leaks, how fast, at what cost. https://t.co/KybX7nQ0mY https://t.co/x5mPk7b5aB

Jevを提供しているType Safe AI のサイトのAPIが使えるようになり、Vercelの無料の制限から解放されました。 そこで今度は、音声会話から任意の部分で区切った各文章を質問、意見、事実、その他、不明などに Jev で判定してもらうアプリを作ってみました。 音声の文字起こしに有料APIを使うのは勿体ないので、Windows標準アプリの音声文字起こし（Win+Ctrl+L）を使い、そこに表示された文字を使うことにしました。 Windows標準の文字起こしからはUI Automation経由で文字を取得するようにしてます。 LLMのクレジット節約のため、このアイデアは我ながら良かったと思いました♪ ※画面下側はWindows標準アプリのライブキャプション 音声はYoutubeより

released hedos v1.4.3, you can now run @typesafeai's jev like models, laya by https://t.co/Pxj2Wc5iUD for example. hedos serves them on typesafe's own systemone endpoint, so a client written against jev reaches your machine instead. https://t.co/3ISeyitypj

I got a bit of JEV FOMO and wanted a real use case for it in Hopp. I decided to prototype auto-zoom for screen sharing. @typesafeai’s JEV picks how far to zoom from app names and window geometry. When you pair on an ultra-wide, the viewer zooms into the focused app and widens between windows. In a one-hour pairing session with 10 JEV calls per second, I spent about $0.02 on JEV.

Jev + GrokBot is the best GTM system I've ever built It just made my GTM x200 CHEAPER and x400 FASTER than what 95% of teams are running... setup takes literally 9 minutes: prompt → GrokBot → Jev routes every candidate → GrokBot opens only survivors → campaign step 1 → open @typesafeai, create API key (keep it off chat paste) step 2 → tell GrokBot: store TYPESAFE_API_KEY in the secure field step 3

Gave Jev by TypeSafe a spin in TurboCode. Jev has gotten a lot of attention recently for being a fast classifier, or what TypeSafe calls a 'System One Model', taking state in the form of text and returning booleans, numbers, strings from a collection, etc. Short demo below. https://t.co/e1gMheNTM8

Had to try Jev, so I built a mood tracker where your writing stitches a flower. 🪡I🌸 As I type, Jev scores each sentence against 6 feelings. The strongest supported feeling colours a section of the flower. 🫠 @typesafeai https://t.co/FSFZgxL6Zw

@notkevinzhang Built CiteGuard — dump a poisoned RAG corpus and Jev Keep/Drops the citations so your agent doesn’t cite junk. https://t.co/tp6roerfnz https://t.co/EmxStapj4p

Got off the waitlist for TypeSafe’s Jev early access and asked my agent harness for a hands-on demo. One starting prompt (multiple agents and review cycles under the hood). Astra plans and reviews; a Gemini swarm builds. The result: a Minecraft factory sorter with Jev deciding where items go. Demo: https://t.co/KI7l3OO6nK Code: https://t.co/WiWjj89Hqm

my ai agent now has a f**king conscience. it costs $0.001 a decision. every time it wants to act, it stops and fires one Jev call first: act or hold. Jev answers in milliseconds for almost nothing, so i can gate every single move and the cost never blows up. here is the exact setup: > agent proposes an action > Jev scores it on 6 things: reversible, in budget, backed by data, downside capped, ever

jev is insane it makes a landing page audit cost less than a rounding error. built this with @typesafeai > i paste a url > it pulls the headings, copy and ctas > jev answers 30 typed questions in one pass > every score opens to show the probability behind it cost: $0.00024 per page time: ~1.1s per page i pointed it at stripe. it scored 81:

First experiment with @typesafeai's Jev A semantic log filter. (with a CLI that lets my agent use it too) Ofc, I had to overdoit and build a TUI around it :) https://t.co/wG2OrPgaxG

Built a Rubik's cube solver with @typesafeai's Jev 🤓 Jev picks every step, live, with a graph showing the stickers sliding along their orbits. No solver library is involved. Jev doesn't know cube algorithms either. The trick: code generates legal steps for the current stage and describes what each one does. Jev only judges "which of these 12 is best." 17 decisions later, the cube is solved. The co

JEV maxing of course, but this time drinving a browser for Hermes agent (--tui with https://t.co/5zES5g85zi) and with minimal token ingestion. Each step: DOM → element table → Jev picks click/type/select AND whichelement, one OpenRouter call. No screenshots, ~200 bytes back per step,~$0.0001.

You manage 1,000 customers. Monday morning is coming. Who actually needs you? I built an experiment with @TypeSafeAI's Jev to find out. → 1,000 (synthetic) SaaS accounts → 6 judgments per account → 6,000 semantic judgments → 9.37 seconds → $0.0775 → 180/181 planted issues caught The interesting part isn't the dashboard. Jev never decides who gets prioritised. It judges six narrow questions — atten

Tried @TypeSafe ’s Jev for a CEO Barometer experiment on S&P 100 companies. Jev looks like an interesting tool for data analysis, particularly in terms of speed, output quality and low cost. Curious to explore what else it can do. https://t.co/nOBddaDW9g

Jev made 973 decisions in one agent run and generated 0 output tokens doing it same run, same output: $22.18 on a reasoning model before, $0.05 after, and the frontier model was never asked once the loop: prompt → structured state → Jev picks the model → worker runs → Jev scores the result Jev gates the tool call deliverable each decision ships a small state object, not the transcript. that is why

JEV IS INSANE. I built a real-time trading bot in one evening and one morning. It reads onchain data, offchain data, social noise, liquidity, wallet movement, all of it. Then it makes trades by itself. Fully autonomous. So far it has lost me $31,680. The funny part is that it works. It is fast. It does not hesitate. It never gets tired. It takes every rule I gave it seriously. The problem is that

I combined ChatGPT browser + JEV for job portal applications. The result? 2-3 Times faster results compared to Browser mode only https://t.co/y6vP9bHXSf

I had Jev play @poncle_vampire's Vampire Survivors. It worked as advertised - zero shot, fast (usually 200ms, over 1s in a slow block of time). I'd love a little more info on why it made a given choice for debugging. This was definitely because of my prompts, but for some reason it did not pick up any items, even ones that would have really helped like the empty tome. I wish I could go back and se

Still having fun with my Jev movie classifier (📣 sound on) What have you done with Jev so far? 👇 https://t.co/JMikbbt55T

I built a racing game where Jev is the driver with Claude Code! Jev picks the racing line and when to brake, given the speed and the corner ahead. Jev has no memory, so after each lap the game hands it the times for all 11 corners and it decides where to push harder. https://t.co/sUAqKJZQEt

I got access to Jev, a new kind of AI model. To understand it, I built a small test. GitHub marks 321,709 beginner tasks as "open, nobody working on it." I had Jev read 290 of them to check. 5 seconds. GitHub says they're free. 77% are not. Why Jev is useful: most AI models talk. Jev decides. You give it text, ask a simple question, and it answers yes or no, plus how sure it is. That makes it fast

I used Jev to build a universal search for my browser. Introducing ⌘ Cmd-F. - ask any question on a site and get answers instantly - can also retrieve cross-page results - 100% free (pay only for Jev API) - completely open source Never using find again. https://t.co/joVST2yW6p

Jev用マインスイーパーが、あて先をローカルのjev互換API対応llama.cppに向けるだけでちゃんと動いた！ あと、Qwen3.8-Flash-Nextがマインスイーパー解いた！ 動画は9倍速で、実際は4:30かかってる。 https://t.co/74MFaDXYWS

I TURNED JEV’S TYPED DECISIONS INTO A GHOST. Built JEV BOARD, an Ouija-style board powered by TypeSafe’s JEV. The use case was simple: give JEV a structured set of possible next words, let it choose the strongest path with probabilities, then make the planchette physically spell that choice across the board. Every reply is built through real decisions inside the app. The board gives those decision

【15項目の入力が、わずか1.2秒で完了】 面倒なフォーム入力、アカウント情報の自動入力はあるのに「他はなんで手入力のままなの？」とずっと思っていました。 そこで、話題のTypeSafe AI「Jev」を使ってChrome拡張機能を自作してみたのですが……想像を絶する爆速で感動しています。 文章生成をスキップして構造化データだけを返すJev、こうしたブラウザ自動化との相性が抜群すぎる！

Stop making your users click through menus. “Do it for me.” “Bring the controls here.” “Show me how.” I built Waymode so your app can do all three, using its existing actions and UI. Open source. Powered by Jev. https://t.co/qkHQJV3Dzg

I gave ChatGPT and Muse the same prompt “write a plan for a Jev trivia game” Then I fed the prompts to Jev to judge. Here it’s judging muse on the left and ChatGPT on the right https://t.co/6dVnInfN0D

Jev ก็เล่น Tetris ได้เหมือนกันครับ 🤖 จากที่ลองเอามาช่วยโปรแกรมตัดสินใจ รอบนี้ให้มันเลือกแผนวางบล็อกในเกมที่เปิดบน Browser บ้าง เว็บที่ใช้เล่นก็คือ https://t.co/yqYtnIVsCL เนาะ เนื่องจาก Jev เองไม่สามารถอ่านรูปภาพได้เราเลยต้องช่วยมันนิดหน่อย 👓 เบื้องหลังคือเราแบ่งงานกันแบบนี้ครับ 1. รับภาพจาก Browser แล้ว crop เฉพาะพื้นที่เกม 2. ให้ Python อ่านสีพิกเซลตามช่องกระดาน 10×20 เพื่อแยกว่าตรงไหนเป็นกองบล็

拿 Jev 做了个安全相关的小实验：速度真的快！命令还没敲完，旁边就开始提示风险。 把 rm -rf ./build 改成 rm -rf /，提示也跟着变了。 又找了 200 条带标签的命令跑了一遍，结果放在视频后半段。 https://t.co/kTOcaGURAk

@CompleteSkeptic Diogo, you may not have been around for the 1981 arcade era, so I fixed that. Meet Jevlaga: 80s arcade vibes, a deterministic game engine, and your baby Jev clearing mobs. 👾 Works pretty well 🩷 https://t.co/hXB6jbc1ud

JevでAI臭チェッカーツール作りました。 ・Codex：企画・文章・実装 ・Jev（typesafe-ai/jev）：文章のAI臭を数値化 ・HyperFrames：画面制作・動画書き出し ・FFmpeg：MP4化・最終チェック ・TypeScript／Node.js：処理の実行環境 AI臭の数値は演出ではなく、Jevで実際に計測しています。記事は👇️ https://t.co/Y9oalGWewM