Jev for browser agents — a daily feed

AI ハッカソン #AIHack #OrcaRouter 8チーム目は > AI Integration with Jev Agent > ターミナルベースのAIラッパー「Computron」！ 先週出たばかりの Jev 💪✨ ターミナルで自然言語の指示を受け取り、Chrome上のタスクを実行します。爆速！安い！ https://t.co/ricznDVbvY

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 https://t.co/jStD5wcjp5

Hello Everyone @X I build FormPilot: Chrome extension that lets Jev fill the form. We One click. Green = write. Yellow = you check. Page → scan → Jev (which key? which value?) → confidence gate → fill. Your key. Almost free per form. Reply pilot or DM — I’ll send the repo. Next up from the same bench: • more Jev agent architectures • small tools that decide, not chat #Jev #BuildInPublic #software

Agentic browser testing might finally be solved using Jev. I built an open-source CLI + skill that lets Claude/Codex test frontend features in a real browser using Jev, then open a PR with screenshot proof all by itself. Demo run: ~7 seconds. Tiny cost. https://t.co/7a8OEnstA7

@typesafeai 's Jev in your browser 100 requests cost me $0.005 https://t.co/btthsyCrwp

Just made a simplified web interface for AI decision making models like Jev. It can: - Choose the best option - Score decisions - Give Yes/No decisions Supports plain text context with a simple interactive UI, built to make Jev's decision making capabilities accessible to casual users.

A normal AI agent needs ~20 clicks and ~150k tokens to book this flight. Jev + WebMCP did it in 6 sentences, with under a second of model time. And it still stopped and asked me before paying. How it works 🧵 https://t.co/TrqbD9QkGX

I let Jev drive my gpui app ui, basically browser use for GPUI. it's not perfect requires gpui-pre 0.3.5 and a patch, but has been very effective for me in agents testing my gpui apps. https://t.co/R3K3uOogvL

Built Yap over the weekend: talk to Chrome, it just does it. Runs on TypeSafe's Jev. A nightly Claude routine reads telemetry and tunes the prompts and tests. This video? Opus 5.5, one prompt + a few tweaks. Self-learning apps are now hands-off. https://t.co/wc6Z4npTRs https://t.co/VhvdrQ6GHD

基于FastJev 跑通了 Jev Ultrafast 的官方浏览器任务。 单张 RTX 5090，本地 Qwen3.8-27B EXL3 连续完成 Lisbon 搜索、Design 筛选、免费取消和打开 Casa Flora：5 个浏览器动作，第 6 次决策返回 DONE。 单次录制端到端 10.375 秒，决策请求延迟中位数 1.388 秒；FastJev 0 输出 token，录制中 0 次 TypeSafe API 调用。 源码 revision、决策 trace 和录像都在这里： https://t.co/S7SpS13066

using hyperspace agentic-os running on my macbook + jev + a frontier model to order me the zero to one book on amazon. ps: this video is condensed for posting here, actual automated drive took ~30 seconds. cc @JeffBezos @elonmusk @CompleteSkeptic @buccocapital

I created the worlds fastest typescript browser agent with @typesafeai for Partyline (The worlds best imessage agent for a good time: +1 (650) 665-4888) It matches @browser_use Ultrafast and is 100% opensource. https://t.co/1R4XFrBfFZ

Jev launched last week to a lot of hype. We put it to the test and made our browser-use agents 3.5× faster over the weekend The experiment reinforced a broader view: fast, reliable agents will combine frontier reasoning models with specialized models 🧵 https://t.co/ZD9vKkm8oB

On very basic tasks, like "open settings, enable dark mode, and save", Jev succeeded 42.9% of the time It routinely fails on tasks such as multi-field forms, spreadsheet-cell editing, rich-text formatting, drag-and-drop, embedded forms, shadow-DOM controls, and file uploads https://t.co/HI6R0cEPJM

@maestrooth @typesafeai Love the Jev-decides → Bot-acts loop. I packaged the browser side the same way: Jev drives watchable Chrome; screenshot CU is the fallback when the action list is empty. https://t.co/R9QM9WAObj

I disappeared from X for a bit. I was building. Jev? Déjà vu. Browser Use + Jev: 7.1s Google Flights. dejevu + plain Llama 3.3 70B: 5.6s. 10 model calls vs 17. 5.6× fewer tokens. 1 API key. No daemon. Every run code-verified. Traces public. https://t.co/oQ6IhcRhjE https://t.co/WwYpu1U8nU

here's my jev use case tightened up a bit today. worked through a lot of edge cases navigating phone systems https://t.co/7UaYi8uSAB

how it works: jev picks each action from the elements actually on the page. code does the clicking. a step passes only when a fresh read of the page proves it. the model saying "done" is not enough. the parts i care about: → runs on localhost, in a private chrome → passwords never reach the model → delete / pay / email actions stop and ask me first → every step: verdict, reason, screenshot the dem

受 Marcus Lowe 那条爆火推文启发，我给 Omarchy 做了个开源插件 Smart Paste： 🔒 只用你复制过的内容，不读屏幕，密码框永远不碰 ⚡ 决策由 Jev 模型完成，命中缓存时按键即填（约 0.4 秒） 🐧 目前只做了 Omarchy 这一版：Hyprland 按键 + Quickshell 状态栏 📖 开源 MIT，一条命令安装 https://t.co/Xbs7fk8b8f

THIS BROWSER AGENT FOUND FLIGHTS IN 7 SECONDS FOR $0.0039 not because the writer got smarter. because Jev picked the next click instead of another LLM essay. writer stays on the writer side. gate stays on the gate side. if your loop still lets the agent say "done", you are running the slow version. watch the clip, then read the breakdown below https://t.co/1m3KoCASjC

made this for myself. no idea if anyone else needs it. go ahead, tell me why it's dumb while you just scroll over the feed on Jev 😒 https://t.co/V8WJECPAia

我做了一个 JEV-UltraFast 的免费开源版本，叫 KaLM-UltraFast，完全在本地运行。 KaLM-UltraFast 负责选择浏览器操作。你可以查看每一个决策。 没有托管的 JEV API。没有按操作计费的收费。 代码 + 演示： https://t.co/XLIW0oE5AZ

1,092 BROWSER CALLS. THEN JEV CUT IT TO 101. Same task. Same browser. 91% fewer calls. → the model doesn't see the whole page anymore — just a short, fresh action table → pick operation + target, that's the only decision it makes per step → the small LLM only wakes up when a field actually needs typing 25% lower median time. Not from a faster model. From asking it fewer questions. Zürich → London,

插件已完全开源，纯客户端运行：Key 存在本地 https://t.co/FPi4abgjlg.local、不碰私钥、不读 GMGN 登录态。 羊毛指路：Vercel AI Gateway 上的 typesafe-ai/jev 标注免费到 2026-09-25，用自己的免费 Key 就能跑。 它不是暴富外挂，只是一副过滤一眼假垃圾的“战壕护目镜”。 源码 & 安装包自取（欢迎 PR / Star）： 🔗 GitHub: https://t.co/DWgLOoPLU8 📦 v0.1.6: https://t.co/ZDKX4rcPLw 觉得有用欢迎转评赞 🔁 DYOR。 有问题或者需要支持什么功能，欢迎提issue或者在本推文下面评论

A browser action as a choice from a list, not a string the model writes. AgenticOS's browser_choice shipped in v0.0.479. Jev picks the operation and the target, reached through Pydantic AI's typesafe extra. Asked it for two Wikipedia pages and it ran two browsers at once. 🧭 https://t.co/QtsJvzDolW

This browser agent runs on Jev and does multi-step web tasks in seconds, for a fraction of a cent. It's called Jev Ultrafast. Give it a URL and a goal in plain English and it clicks, types and selects its way through the site. https://t.co/rHiD0fagpa https://t.co/rtcqnlFvkT

Laya now runs entirely in a browser tab. Not the UI — the model. Free alternative to Jev. Context in, typed answers out, milliseconds. No install, no key, no server. 8-bit ONNX, 524MB, onnxruntime-web over WASM. Cached after the first load. https://t.co/UaFaBky76B

Jev has been blowing up lately. If you've got the Jev API but don't know how to play around with it yet, you can just copy this checklist. 1. jev-ultrafast A high-speed browser Agent built with Browser Use. Jev only judges "what to do, which element to click" at each step, and only calls the small model when typing is needed. Searching for a flight on Google Flights takes about 7 seconds. https://

Jev is the most exciting AI release in months. So I built the open, in-browser version 🎉 open-jev: System One style decisions running 100% on your device. Text in, typed answers out, one forward pass. Nothing is generated, so it can't hallucinate. https://t.co/u4PtKWUmGr

話題のJevを使って、ブラウザのフォームの入力欄を全自動で入力するChrome拡張をAIに作らせてみました。 いやー、ホント速すぎてビビる。 1フォーム入力あたり0.024円。 応答時間は245〜460ms これはメッチャ使えますね。 リプにChrome拡張ファイル置いておきます。 https://t.co/PS183NMg5m

Jev really inspired me to build SUPER fast browser agents without sacrificing accuracy. I gave Codex access to vLLM on 2×B300 and let it change everything from the harness to inference. The constrained optimization: > min end-to-end task time > s.t. score ≥ baseline Caching, thinking, action batching, inference. Any part of the pipeline is fair game. First results below on 12 local form tasks. The

I turned my Jev voice control app into a chrome extension. It can: - search - navigate - click on links - scroll - open tabs - and a bunch more Not sure how useful, but if people want this I'll put it on the chrome web store https://t.co/gVxFQzIDEw

A tiny open source browser agent using Jev instead of an LLM for every click. Found a flight search in 7 SECONDS. Total cost: $0.0039. Here's why that's not a typo. A normal browser agent asks a chat model "what should I click" on every single step. That's a full generation call, just to pick a button. Mine doesn't. The DOM state at each step becomes the input. Jev gets the available actions as a

6、jev-browser — LLM 规划，Jev 负责执行阶段的快速浏览器决策，支持 CLI + MCP https://t.co/fFbF2GQCd1

JEV-BROWSER After testing and learning Jev last week I've concluded that the most useful use-case for me would be to improve the way LLMs interacting with websites. https://t.co/OLLEdIlUTe this is what I'm using for my claudes/groks/devins to do things on websites for me It's a COMPLEX tool and they often ignore the --help and revert to their shitty eval() and writing scripts instead of using quic

@aiversedesign Same idea, one step further: it doesn't just pick the model, it fills the screen. I type a sentence and the right widget is already filled in. Tap Save. ~430 ms, $0.0005, nothing generated. An LLM authors each tracker once. Jev runs every sentence after. https://t.co/YWjywD1MDQ