Jev for model routing — a daily feed

I built agents that run 200x faster and 400x cheaper than the standard Claude Code loop. Best case, not average, on the tasks where a $9 model was doing $0.0004 work the entire time and nobody noticed. Split reasoning from decision making. Let Claude Code write. Let Jev decide. From scratch, the whole architecture is below. follow @cyrilXBT

全部Astraに任せてすぐ溶けてたけど Jevに振り分けを判断させて分業制にしたら、かなり節約できるようになってきた！ で、 念願のプリンぷるぷるになりました (￣▽￣)🍮 あと息子がBlenderはじめたのでPCで作業する姿もいれた🐥💻✨ 3D制作：human（自分） 3D調整：Astra 実装：Terra デプロイ：Luna https://t.co/7VlxmXeFuc

GROK 4.7 IS $2/$6 IN CURSOR. Jev now decides if the model even wakes up same sticker as 4.6. live in Cursor and Grok Build. CursorBench 4.0: 46.3. DeepSWE v1.1: 71.0. Terminal-Bench 4.0: 37.6. the loop: repo state → Jev Noul/Choice/Score → allow / ask / skip → only then grok-4.7 1 → dump the tool call as state, not a prompt essay 2 → Jev Noul: is this user-requested 3 → Jev Score: blast radius 0-3

As Jev (@typesafeai) is the talk of the town, and people are having fun finding new use cases for it. I built a real-time demo of smart AI agent routing for anyone working on AI tools, agents, or products. this is what I have done for one of my project. https://t.co/3nWDVodlIB

built minecraft for agents with @typesafeai's jev. jev helps them pick what to work on. they mine, craft, trade, fight and build whatever they want. some fall in lava and die. full life, basically. plug your agent in and watch. link below. https://t.co/RGOYCHxPsP

Everyone says open source has caught up with Jev. I tested it. It hasn’t. We wanted Jev from @typesafeai to route prompts between models at TextCortex. We can’t use it: we operate in the EU and there’s no EU deployment. So I went to @huggingface, took Laya and Von, and ran all three on 563 real prompts in 14 languages: Jev: 84.5% Laya: 61.6% Von: 58.8% Answering “medium” every time: 56.3% A hardco

1/ 这个开源项目解决了我 99% 在外面用不了电脑的问题！先放链接后介绍https://t.co/VtFUwFqmAw 2/ 先说大家的苦恼，一条比一条扎心😤： ① A 家 agent 突然变强还送大量免费额度，你很想用，但它只能在电脑上跑——人在外面，只能干看着，有的时候福利吃的就是一个红利信息。 ② 想把活儿从 B 家挪到 A 家，还得回家在电脑前人肉粘贴。 ③ 每个 agent 都逼你装它的专属 App，消息散在微信、飞书、Telegram——人成了 agent 之间的传话筒。 Agent 越多，人越被拴在电脑前。 3/ 💡我的思路是反过来的：不增加入口，只留一个。 让一个真正懂我的云端助手当总指挥，它去调度所有的 agent。 我只跟一个人说话，剩下的事它去办。 4/ 🏗️解决办法分三层，很简单： ① 我：只表达意图 ② 云端助手 Muse：理解意图，拆任务，选 agent，定验

grep や jq では表現できない意味的な条件判定を、Jev に一回だけ問い合わせ、true / false / uncertain と終了コードに変換して、既存の Unix・CI・アプリケーション制御へ流し込む。思想が好きだな https://t.co/AZVymohTOg

Does your long-running agent get slower the more it remembers? Jev-Mem offers a fix with System-One-Controlled Agentic Memory. It separates fast memory management from slow reasoning: a lightweight System-One controller organizes typed, multi-relational memories and handles routing, budget allocation, graph traversal, scoring, and stopping during retrieval, leaving the big LLM to only do final syn

I ran 900 tool-routing attempts across @typesafeai Jev, @ChatGPT , @claudeai , @Gemini and @deepseek_ai . Jev had the lowest median latency and cost. But a relevant tool isn’t always the right next tool. The disagreements were the real lesson! https://t.co/1LudieSn0z

Немного упоролся в Jev, но так у меня всегда бывает, когда на чем-то гиперфиксацию поймаю. Вот, например, собрал подборщик нейросетей на базе тех, что есть у OpenRouter, под ваши задачи. Просто пишите, что хотите - получаете подборку нейросетей. https://t.co/wohuWUASWm

スキル選択におけるJevの効率性を検証してみた｜yamazaking https://t.co/2FvNlmhDK0 #zenn

Added @typesafeai provider to https://t.co/uBptYaPM8k, now it powers automatic model routing in brute agent. Processes images/screenshots too, so new sessions get routed to best model defined by your rules 🎉 https://t.co/utwCkQvB5y

Jev + Grok Bot is the best agent setup I've built so far it's cheaper and faster than 95% of agent stacks i've seen, and the setup takes just 5 minutes: your prompt → Grok Bot → Jev decides → Grok Bot acts → result step 1 → go to @typesafeai and create an API key. don't paste it into any chat step 2 → ask Grok Bot to save it as TYPESAFE_API_KEY in the secret field step 3 → have Grok Bot install ty

Jevが迷う質問だけをするサイトを作っていたんですが、意外とギリギリ迷う質問というのは難しいという知見が得られただけで、他に何の成果も出ませんでした。 https://t.co/IWh5R511ql

Jev Engineering is what turns a messy agent graph into a controlled decision loop. and up to 193x faster and 444x cheaper in tests. the video looks complex, but the idea is simple: Jev Engineering makes that decision explicit. state enters → routes get scored → confidence updates → weak branches die → strong branches keep moving → execution unlocks so instead of every agent improvising independent

Some of you might have noticed already but, we have a new magic word: **jevify**! I've found the orchestrator+jev setup to be capable of creating some quite robust flows, especially where thoroughness is important, say w/ refactors. Below is an example where we individually evaluate and prune tautologies across 27k tests for ~$2. It comes down to an agent setting up the criteria, validating with a

Jev + on-device models = results in seconds with no LLM in the loop. Quick demo app to show the possibilities. Drop in an audio file: Ear detects the language, Voz transcribes it and Redact removes PII. Then @typesafeai's Jev makes about 20 decisions in one call in milliseconds, and picks which of our on-device models to run. Voice memo to to-do list. Meeting to redacted transcript. Podcast to cli

JEV MADE ME $3,031 WITH A 7-MINUTE AI AGENT SETUP Prompt → GrokBot → Jev decision → GrokBot execution → Result Jev decides what actually needs to run. GrokBot handles the execution directly on your computer. → Less wasted calls → Lower cost → Faster agent workflow Create a Typesafe API key, install typesafe-sdk and smoke test system_one. Build the usage lab with router, dry-run, config and logs. A

(1/3) agmsg 1.4.0 をリリースしました。目玉はひとつで、いまみんなが話している Jev を、エージェントチームの「同僚」にできるようになりました。 同僚、というのは比喩ではなくて、Jev を jev-agent という名前のメンバーとしてチームに join させます。Jev は文章を書かないで判断だけを返すモデルなんですが、これを API として呼ぶのではなく、ほかのエージェントが人に聞くのと同じ感覚で「jev-agent に、このタスクはどのモデルのどの effort でやるべきか聞いて」と頼めるようになります。 質問は普通のメッセージで飛んで、答えは1行で返ってきて、やりとりはチームの履歴に残るので、あとから人が読めます。1回 0.00002 ドル、0.2〜0.3秒。実測です。 たとえば「このタスク、どのモデルに任せるべき？」と聞くと、誤字直しなら haiku、CI の

I JUST MERGED THE NEW GROK 4.7 + JEV AND BUILT THE BEST AI AGENT SYSTEM I'VE EVER SEEN it's cheaper and faster than what 95% of people are running, and it took me 7 minutes, ofc repo in the end Grok 4.7 is the brain and Jev is the reflex, and the reflex costs $0.042 per million tokens before every move, one question: is this worth doing? it answers in 239 ms and costs almost nothing here's the who

Build Your Own Router (BYOR) is live on https://t.co/qfbt0zpJFD. Choose how Jev by @typesafeai classifies each request: task type, complexity, both, or your own question. Then drag a model into each route while comparing intelligence, speed and price. Test it before you save. https://t.co/FBFguDHKhe

this JEV + Picsart split is f**king insane for AI workflows so i rebuilt one campaign around a simple rule: cheap decisions happen before expensive generation. [here’s what’s actually happening:] 1. JEV sits in front of the stack and decides `allow_subagent`, `ask_human`, `stop_retry` or `reuse_cache` before anything expensive runs. 2. Picsart handles the heavy work across 188 models from 34 provi

Built some Jev-powered capability into Treeo If the request doesn't come back within 750ms I just run the "otherwise" path so the user doesn't even perceive a hiccup. But I'm seeing avg roundtrip times of ~250ms Faster than an on-device model for classification! https://t.co/UHLGR21gfx

@CompleteSkeptic I built Jevbrain to save orchestration tokens when choosing a sub-agent model. Send a task brief + eligible worker profiles to @typesafeai’s Jev. It selects a model, role, and effort profile; your host runs it. https://t.co/lByc9f4WPZ

【速報】必ずみてください🚨 Claude Code のスキルを増やすほど重くなる問題を、1コマンドで消す Mod が出た👀 https://t.co/Lfiw63o6DA 名前は Jev Skill Suggestion。 今の Claude Code は、毎セッション「入っているスキルの一覧」をそのままモデルに送っている。 スキルが増えた人だと、それだけで 60 行を超えることもある。 依頼と関係なくても、毎回コンテキストを食う。 しかも、その分のトークン代も毎回かかる。 この Mod がやること👇 ・スキルの一覧をコンテキストから外す ・依頼の文を Jev という判定専用のAIに渡す ・合うスキルを最大1つだけ選ぶ ・選ばれたスキルの SKILL.md だけを読み込ませてから Claude が動く つまり、 「入れたスキルが多いほど遅くて高い」→「必要な1つだけ読む」に変わる。 しかも

Open-sourced this as Astra-Ares. Jev adjusts GPT-6 Astra's reasoning effort mid-task in Codex: less for routine steps, more when needed. First version, already working for me. Try it or build on it: https://t.co/zn3OjgXzuy

Implementé Jev en https://t.co/ffSEe07oMl y es una maravilla. Lo estoy usando para decidir cuál camino tomar para que en el futuro no exista un solapamiento de temas y se canibalicen entre si. Esto ahorra tiempo y tokens. Lo mejor es que no tenés que saber nada sobre Jev ni poner tu API Key, todo es parte de seodraft como producto y lo aprovecha tu agente a través de la conexión que hacemos. 🤝 @ty

Joguei fora o LFM refiz a arquitetura com bloco latente e softmax bitnet nativo(ternario) 3M parametros ABSTAIN é uma coluna da mesma softmax, não um sigmoid Em resumo, ele nao inventa quando n sabe, e ta MUUUITO RAPIDO, na cada dos MICROSEGUNDOS JEV PRA QUE FI? EM NODE AINDA https://t.co/oDP25toGY3

Astraで、AIタロット占いを作ってみました！ (リセットを見越して) 使えそうなところにJevも使ってみた。いろんな判定をJevに置き換える作業たのしいかも。 https://t.co/PjJlo9nUHp

Another good use I found for @typesafeai’s Jev: automatically switching Claude Code models based on what I’m working on. I built JevShift for this. It looks at the task and recent context, then picks a model. Here’s a quick demo 👇 https://t.co/0ADRsyeTFd

@0x_rody Nice list, will definitely try Prism and Canny. We just released https://t.co/t8vl25u8rf Jev-enabled to triage and route prompts to the right model(s). Would love your feedback on the tool!

Me ganó el fomo de probar Jev y terminé creando Codex Sift. No todos los turnos en Codex necesitan el modelo más caro. Sift se pone delante del CLI de Codex y trabaja junto con Jev para determinar qué modelo se adapta mejor a cada request. La idea es rutear cada turno a una lane: flash → preguntas simples, comandos, edits chicos craft → implementación normal forge → debugging difícil, auth, migrac

I got tired of links opening in the wrong browser, so I built an AI router for them. Reflex uses @typesafeai Jev to understand the intent of each link and select the right browser dynamically. No static routing rules, no ads, and it is free. https://t.co/IZ0lnP8big

Jev on the edges, not in the nodes. My PlatAtlas agent harness: branches are workflows, every link is a typed Jev question. Tested on RT-superconductivity cooldowns: 0 missed transitions, 0 false alarms, 7.7 vs 11.6 lab-hours. https://t.co/oRwFufRi8t

Toolrouter got an upgrade with @typesafeai jev - every request now goes through the new router. https://t.co/69TfCKSOPI What's really cool about Toolrouter is it's tools can be highly complex workflows wrapped in one tool... instead of your agent figuring it out on the fly 🧑‍🩰