X posts

THIS IS INSANE For computer use agent, I have been thinking of what is the best demo case of agent is capable controlling the computer as the human Finally, the best one I found is playing a REAL game, for example the king of fighters, from STEAM on Windows This requires almost ZERO latency if you want to win. So I built one using Sai @sai_borg leveraging @typesafeai - Pretrained a few skills/move to play the game in Sai, including the special move - Sai navigates to start the game and decides which character he is most comfortable playing - Jev decides the best move to make almost simultaneou

Wanted to see if Jev by @typesafeai could make our product's search understand what people actually mean. @FromTheFarmUSA Pretty impressive. Still a lot of fine-tuning to do, but it's already a much better experience. 👇

OODA AI + OpenClaw + VM + Jev = 7,1 sec for a Google Flight Search with Agents running full Computer Use. Uses Sol as primary Model, Luna as Light Model and Jev as classifier. But pick your flavor! We also have Laya and Von and any of our over 80+ AI models wrapped in an /v1/systemone endpoint. OODA AI is an All-in-One AI platform with over +150 AI models across Text, Image, Video, Audio, Avatars. Check the comment for a short video of how we use Jev and System1 models to decrease 50% context use, increase speed on Agent decisions in OpenClaw, handle better intent and tool decisions, Guardrail

Jevが音響監督してくれるアプリ作った！with Opus 5.5 Jevが台本や話した内容 + 過去コンテキストを加味して、曲・環境音・効果音の切り替えを自律的にやってくれる！意外と良さげなユースケースかも？

MY MAC NOW READS MY X FEED FOR ME, 1,318 JUDGMENTS IN ONE SCROLL, 203MS EACH, AND IT TOLD ME 48% OF MY TIMELINE IS NOT WORTH READING. The setup is small (Revealing now, might delete later) ⬇️ > Jev running as the decision layer, typed output only > a feed reader that grabs every post while I scroll > 12 typed questions fired at each post in parallel > is_shill, reads_like_ai, claim_needs_source, worth_reading, mute_author > every answer is a bool, a score or a choice, with confidence > verdict per post: keep, reply, skip, mute The output is not an answer, it is a row {"mute_author": 0.80, "con

I solved Meta Ads Library doomscrolling with Jev Rapid chrome extension Super fast (sub 100ms) and cheap as f*ck Comment your fav Jev use case & I'll send you the prompt

Nirnaya, Jev like decision model, can play subway surfers and dino. I've been playing around to see how far we can push a really fast decision model. Results have been interesting to say the least. Impressive for a model that can fit in a browser for sure.

X驚き屋判定スタンプ 【⚡️生成AI】 パクリばかりですが、今話題のJev驚き屋判定 結構いいのでは？ 今のJevは画像は見れないので、画像も見れるようになるとなおよし 角が立つので自分のポストだけでテストしていますが、「損益自慢！」や「詐欺！」なんてのも出ます。

imagine all your tiny cars parked quietly in your pocket... so I built this for my hotwheels' collection with Jev by @typesafeai - Jev reads your query into intent + make + colour and scores every car 0-100% on how well it fits - Type the name or the serial off the back... Jev matches it against the whole collection, works out if it's mainline, silver or premium and says buy or skip against that card's usual price - Color based sorting: measuring the photo pixels lied (it saw the blue card, not the car) so colour comes from reading the toy itself... "blue with red base" = blue, "pinkish red" =

🧠 $JEV OpenJEV is a free API for the Jev model, which delivers structured decisions (selection, scoring, probability) for software and applications in milliseconds It focuses on enabling rapid, structured decisions within applications—such as ticket routing, urgency assessment, agent tool selection, search relevance, moderation, and determining whether human escalation is required. The model accepts the current state plus a set of questions across three types—Choice (selecting from options), Score (rating on a scale), and Noul (probability that a statement is true)—and returns ready-made answ

Benchmarks suck. So I made Jev and Laya play chess instead ♟️ Insomnia project. Repo in the thread

YOU’RE PAYING FOR AI IN THE WRONG PLACES. The costliest mistake in the AI ​​business is forcing a powerful model to make every single decision. Filter. Evaluate. Select. Weed out. Route. You don’t always need the most expensive model for that. That’s why the Jev + DeepSeek + GPT + Claude combination looks so promising: a low-cost layer handles simple decisions, while expensive models kick in only where they truly add value. The same principle applies to content creation. Instead of juggling dozens of disjointed AI tools, there’s @Picsart - where 188 AI models and tools for generating and editi

Don't let a pretty preview make the decision. The preview says record_A. The action targets record_B. Open-Jev-27B-v1.1 catches the mismatch in this case. Two real predictions. Seven choices. Edited replay; no browser actions. Model, code + data:

This 29-second clip is the entire thesis of my Jev + PicsArt article Take a performance people already recognize. Rebuild it as Minecraft. The viewer understands the idea before the first line lands. Then turn that format into a system: → Jev finds the source, angle and hooks → PicsArt rebuilds the visual world in different styles → you test the strongest versions while the format is still moving One strong idea can produce 10 different clips without feeling like the same post copied 10 times. I mapped the full workflow in the article below. @Picsart

有人问 Laya 和 Jev 的准确度到底差多少，我自己又跑了一轮小测试。 这次还是用 Laya 英文版，一共 100 道判断题： Laya：92 正确，8 错误 Jev：100 正确 在这组测试里，Jev 高出 8%。 当然，这不是完整 benchmark，只能说明在这批偏“理解和判断”的题目里，两者表现有差异。 我主要测了几个容易拉开差距的场景： 1、属性特征描述（Feature Alignment） 例： The sweet red fruit with a leafy green cap on top. 不直接说名字，只给特征描述。 测试模型能不能理解“红色 + 甜 + 顶部绿色叶子”这些条件组合，而不是只匹配关键词。 2、单重 / 双重否定（Negation） 例： Skip the grapes and citrus, give me the other one. 很多小模型容易抓住高频关键词： grapes、citrus 但忽略前面的 skip。 这里考验的是模型能不能正确处理否定和排除条件。 3、上下文干扰和目标指代（Mention vs Intent） 例： The recipe picture has grapes, but the fruit to serve is orange. 前面提到了 grapes，但真正要选择的是 orange。 测试模型能不能区分

I gave Jev a paddle. Can you beat it? 🏓 JevPong is a real-time game powered by @typesafeai decision model. Watch Jev decide, beat it to 7 points and top the leaderboard. Built with GPT-6 Astra in Codex. Routed through Postman Fabric Gateway. Play:

Jev ha estado explotando últimamente. Si tienes la API de Jev pero aún no sabes cómo experimentar con ella, simplemente copia esta lista de verificación. 1. jev-ultrafast Un agente de navegador de alta velocidad construido con Browser Use. Jev solo juzga "qué hacer, qué elemento hacer clic" en cada paso, y solo llama al modelo pequeño cuando se necesita escribir. Buscar un vuelo en Google Flights toma unos 7 segundos. 2. fast-jev-compaction Compresión de contexto para Claude Code. Antes de cada llamada a herramientas, haz que Jev juzgue si hay algo aún útil; elimina lo inútil y mantén el texto

Jev 居然输给了一个 1.88B 本地模型？我用俄罗斯方块重新测了一遍 当我看到 this-that-model-1.0 在 68 道决策题上做到 94.1% 准确率，而 Jev 是 76.5% 时，我的第一反应是： 这是真的假的？ 一个只有 1.88B 参数、可以在 Mac 本地运行的小模型，真的能在决策任务上超过 Jev 吗？ 所以我做了一个俄罗斯方块，让两个模型使用相同的棋盘规则、随机种子和方块顺序。程序负责计算所有合法落点，模型只负责决定方块应该放在哪里。 1️⃣ 一开始，Jev 的表现更好 最初，我先过滤掉明显更差的落点，再把剩下的多个候选位置同时交给模型。 在这一模式下，Jev 的表现更加稳定。 我观察到的一轮里： Jev 消除了 5 行，this-that 消除了 3 行。 如果只看到这里，很容易得出结论：本地小模型还是不如 Jev。 但我后来意识到，这种问题可能并不是 this-that 最擅长的形式。 2️⃣ 换成“二选一”后，结果反了 this-that 本质上是一个 typed decision model，更擅长候选明确、边界清楚的选择题。 所以我把决策方式改成了淘汰赛： A 和 B，哪个落点更好？ 胜者进入下一轮，继续和其他候选位置比较，直到选出最终落点。 换成这种模式后，我观察到的一轮结果变成了： this-that 消除了 7 行，Jev 只消除了

Laya just launched as an open-source alternative to Jev. Built by ConvAI Innovations, Laya is designed for fast decision tasks where you do not need a full LLM to generate text. Give it an email, support ticket, JSON payload, or plain text, and it can immediately return a label, score, or probability. Examples: ▪️ Is this email spam? ▪️ Which team should receive this ticket? ▪️ How urgent is this issue? ▪️ Is this a prompt injection attempt? ▪️ Should an agent take action or escalate? The key difference is that Laya does not generate text at all. You get a structured decision directly, so ther

JEV JUST HIT 18K STARS IN A WEEK AND BOOKS ZURICH TO LONDON ON GOOGLE FLIGHTS IN 7.1 SECONDS WHILE CHATGPT OPERATOR STILL NEEDS 3 MINUTES AT $200 A MONTH jev-ultrafast splits every browser decision into operation, click target, and typed text, so one llm request replaces five and the agent finishes the task before the page finishes animating [here's the setup i'd use:] 1. install it git clone cd jev-ultrafast uv sync cp .env.example .env 2. add the keys typesafe_api_key → .env openrouter_key → .env (for mercury 2.5) 3. wire the models mercury 2.5 → typing (diffusion, milliseconds not tokens) g

Jev founder Diogo Amogo just put out a PDF on building a Jev harness for coding agents the claim: 200x faster, 400x cheaper the model stopped being the bottleneck a while ago. the harness around it is where the cost and the speed actually live hand this PDF and the article below to your Claude Code or Codex instance and let it rebuild its own setup your agent gets a better harness tonight, and monday looks different than it would have 👇

GPT6 Sol×HyperFrames×Jevで架空の美容ECサイトのブランドムービー作ってみた！！GPT6 Sol極高で作ったけどマジでクレジット減らないw

I built a Sales Copilot using Jev → Listens to sales call live → Tells you what to say next → Helps you follow the script and handle objections → Shows you what stage of the call you're in → Gives you live signals and probability of closing Demo below on a recorded sales call:

a bundling/farm detection model using JEV for all new launches on Robinhood costs $0.7/day, ~200 millisecond per token it works like magic https://t.co/KpY36cSlUf

Real-time Among Us using Jev @typesafeai Weekend Vibecode to learn-by-doing w Jev 🧠 Planner: DeepSeek V4 Flash 🧭 Actor: Jev Jev imo a game-changer for real-time agents, runtime verifiers (will implement in @sealgate_ai), and real-time search. 🔗: https://t.co/6aUkb7XToG https://t.co/ildZrrImFV

We started using Jev to make first interaction in chats feel faster. Median response time dropped from 35.8 seconds to 2 seconds. about 18x faster. Try it now at Motion 👇 https://t.co/vSB2oH9kT7

Found this HypeMeter build made with Jev + Minds in just 4 hours. The idea is simple: use AI to scan NFT listings and figure out whether a cheap NFT is actually an opportunity or just a trap. 20 collections, 898 listings 801 screened → 11 traps rejected → 1 passed. Jev handles the fast evaluation, while Minds works as the AI agent. You can give it strategies in plain language, let it monitor listings and receive daily summaries when opportunities appear. It also checks rarity and previous sales instead of blindly calling everything a buy. Every call is recorded and scored again after 24 h

Over the weekend, our team built a novel agentic document processing system using Jev. The initial results are very promising. Our system completed the same editing task 97% faster than ChatGPT's Word plugin. https://t.co/sZFU4qcgUJ

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

So instead of chasing which posts I should pay attention to, I am using Jev (combine with my original Algo) to have this floating panel Now I can quickly find which posts are most relate to my interest, and probably save ton of time because of dump scrolling, but hey, dump scrolling is fun too,

I asked Jev to predict its own future. My last post (below) got 300k+ impressions, 400+ comments, 101 reposts. I used Jev to categorize every comment and quote to capture public X opinion on Jev’s future. Here’s results: ———— *Note: I created five prediction buckets, plus an “other” bucket for comments that didn’t cleanly fit into any of the five, or didn’t contain any substantive text. I’ve excluded the “other” bucket in the chart below. Also, for those who may wonder about the accuracy of using Jev for this, I made a second labeling pass with GPT 5.6 Sol. Jev and Sol agreed on ~8

Meet an artificial Algernon built with JEV by @typesafe 🤖 It receives data about its surroundings and finds its own way through the labyrinth to reach the flag. 🏁 https://t.co/9xYR5aob5i

Made a thing that lets Claude/Pi cite published IETF RFCs using Jev. Provides over MCP/CLI/Extension with quoted citations and ranked results. https://t.co/GfRnFLqZF2

My chrome extension for youtube built with jev now reminds you what you came for. For all of us, who open YouTube for one tutorial and, ten unrelated videos later, still haven't resolved the problem they came for. Thanks @The1Broom for suggesting this. OnPurpose now nudges you after 10 minutes off topic viewing. 10 minutes is by default, you can change the timer as per your comfort. Link to extension in comments

Live https://t.co/6DhHClRm9u filter made using Jev @typesafeai and @OpenRouter Add category scroll and filter out on the fly https://t.co/rrTngMFugZ

Here is a useful tip for anyone using JEV. Once you extract probability estimates using JEV, you need to perform calibration. But what should be the criteria for this calibration, and how accurate are JEV's predicted probabilities in reality? More importantly, how should we adjust our business logic based on the values generated by JEV? JEval is an open-source tool that helps you discover metrics, recommends them, and displays them across various viewers. It also provides a metric-extraction library and a CLI interface. I've even converted it into a skill so it can be fed into AI agents. JE

The bookkeeping services industry is dead. This weekend, I built a better solution using Jev from @typesafeai. I fed it 34 months of work a firm charged $20,000+ for. Jev did a better job in 20 seconds, for just $0.32 🤯 https://t.co/8lA9h32eOo

Soon as I made a project using Jev, it got outdated. The tool launched barely a week ago!!

We use LLMs for a surprising number of problems that aren't generation problems. Been experimenting with Jev from TypeSafe for tool search and ranking. Wrote about what we tried, what worked, why I think this pattern goes beyond search reranking 👇🏽 https://t.co/tnreET4OEa

Introducing Jevussy 💦 (Jev + Debussy) A real-time piano player using Jev. This is INSANE it actually sounds like Claire de Lune 😂 Check it out here: https://t.co/cvU0t6dCCy https://t.co/7A4USP0cay