Resources for building with Jev

Harry Tandy

@HarryTandy

TypeSafe founder Diogo Almeida: "I want to automate the easy work before the hard work" In this 2-hour interview, he explains how Jev uses choices, scores, and yes/no probabilities to route work inside agent loops The guide below has 3 builds: a skill router, a codebase linter, and a retrieval reranker Watch the interview, then copy the builds below

Ricky Grannis-Vu

@RickyGrannisVu

We rewrote the oldest primitive in programming. The if statement takes a sentence now, and Jev decides. That nested boolean mess eleven rules deep, four AND/ORs, one comment nobody understands anymore? Replace all of it with "if this expense is allowed under the expense policy." Judgment Mode, live now.

Ricky Grannis-Vu

@RickyGrannisVu

Jev gives you a probability. You set two lines on it: above the top one it's a yes, below the bottom one it's a no, anything in the gap runs a third branch. It's just a normal branch, so you can handle uncertainty flexibly.

saturn

@SatOnchain

this paper is f*cking gold. someone made handbook on "how jev can be used as classifiers for evals on scale" anyone who solve this has chance to make millions. Bookmark and read full handbook.

st1ne

@SolSt1ne

Jev Founder, Diogo Almeida, just released an independent working note on building a Jev harness for coding agents Not affiliated with the Jev team, just a builder mapping the loop Section 1 - the split: why harnesses already make these calls, just badly, with regex parsing prose replies Section 2 - five calls to move: files, model, safety, done, context, each one a typed Score, Choice or Noul instead of free text Section 3 - what it saves: ~400x cheaper per decision vs a cached frontier model, 0.2s median per call vs seconds of LLM latency $605 per 1,000 decisions on cold frontier calls down to $0.17 on Jev Bookmark it and read it tonight Then go rebuild your own agent loop

VIctor Janni

@VictorJanni

If you liked the idea of Jev (ultra-fast, typed decision models instead of heavy 70B LLMs), meet Laya—the open-source equivalent. Why it matters for production pipelines: 1. Blazing Speed (<100ms): Avoids slow token-by-token generation. Evaluates options in parallel to make routing/classification decisions in milliseconds. 2. 100% Self-Hostable: Run it locally on cheap hardware (even CPUs). Zero API costs and zero data leaving your servers. 3. Native Typed Outputs: Outputs structured probability distributions directly. No messy regex or JSON cleanup needed. Top Use Cases: • Fast Model Router: Route 80% of easy queries to small models and save frontier LLMs for hard tasks. • RAG Filter: Score and prune retrieved chunks in 40ms before generation. • Real-Time Guardrails: Catch prompt injections before they reach your main agent. The future of AI architecture: Local open-source models handle the fast plumbing. Frontier models handle the deep thinking. GitHub Repo 👇 https://github.com/aayushch/laya #OpenSource #AIEngineering #LLMs #MachineLearning #SystemDesign #BuildingInPublic

Théo

@cryptofanatiix

Most people slap the word “AI” on any Buy / Sell button and leave it there. Jev is a model that came out this week to make fast choices. You give it a situation, and it answers with a decision and a probability: buy, sell, or wait. That takes about 200 milliseconds. It is not a model trained by the Qubic network. It is a tool you call, like an API. X is full of people who wired it to a ticker. Some taped it onto Hyperliquid over a weekend. A few have already posted their losses. That shows what it is: a call returned very fast, then the bot keeps going on its own. So the real subject is not “AI can trade.” It is not mixing two things that do not have the same job. Jev gives an opinion and stops there. It does not stay attached to the error. If the trade is wrong the next day, Jev is not the one carrying the score. The account of whoever plugged in the bot is. BPP-9000 is not built that way. Qubic miners train a small network on a year of Bitcoin hourly candles. The network has to say whether the next hour is more likely to go up, go down, or stay uncertain. Errors are counted. More errors means a worse score. Ant Colony keeps the version that misses less than the last one. That work stays on the network. You can still open a page later and reread the result. The difference fits in two sentences. Jev is a judgment you ask for the length of one call. BPP-9000 is a model the network has to carry for a full epoch. I am not running my bag through a bot someone stood up on a weekend. I am waiting for a public score I can refresh. The question I keep: the day both put up an arrow, which one can you still reread the next morning?

delost

@thedelost

this is pure f*cking treasure 10 GitHub projects that put Jev between your agent and every decision your agent should stop paying a frontier model to answer yes or no ROUTE THE WORK 01 jev-router > https://github.com/gargpratyush/jev-router 02 jcm-router > https://github.com/adarshmishra07/jcm-router 03 jev-agent-skill-router > https://github.com/GodsBoy/jev-agent-skill-router GUARD EVERY ACTION 04 pi-heed > https://github.com/Nyarlathoteppppp/pi-heed 05 fx > https://github.com/vercel-labs/fx 06 Foreman > https://github.com/thruwire/foreman SCORE WHAT MATTERS 07 LlamaIndex Jev > https://github.com/WiktorB2004/llama-index-jev 08 citation-verifier > https://github.com/MarissaFamularo/citation-verifier PUT IT TO WORK 09 yoshi > https://github.com/compozy/yoshi 10 mastra-jev-moderation > https://github.com/CodeAlive-AI/mastra-jev-moderation the loop: your agent hits a fork > Jev answers with a probability > your code decides what that number allows > the agent moves on 3 builds I'd explore: coding agent: jev-router > fx > Foreman > yoshi support desk: jev-agent-skill-router > mastra-jev-moderation > LlamaIndex Jev research: jcm-router > pi-heed > citation-verifier save this, then build a decision layer for your agents ⭣

MIKE

@mikenevermiss

Jev is brand new and there's no tutorial for it yet. so I wrote the prompt that makes Claude read the docs, pick one real use case, and build the agent. it's a runnable code, added failure handling, and a rule for when Claude takes over. paste it in and try it.

darkzodchi

@zodchiii

Send this Jev prompt to Opus 5.5 or any AI agent it installs Jev → audits your workflow → finds where you waste tokens, time and money then moves repeatable decisions out of expensive Opus calls → upgrading your AI stack beyond 95% of people Copy this now. Thank me later

NO1ennn

@N01ennn

x.com/i/article/2103…

Gipp 🦅

@gippp69

x.com/i/article/2103…

Erik Spock Gafni (hiring!)

@EGafni

my personal faves are: AutoResearch for Feature Extraction Use LLMs to generate Jev questions's who's probabilities are fed into a classic ML algorithm like logistic regression to predict a label. Hierarchical Classification Traverse a classification taxonomy using Jev and beam sesarch. Jev is great for graph traversal.

Trackmind

@0xTrackmind

x.com/i/article/2106…

Asteri

@Asteri_eth

Claude + Jev is the first agent setup where I stopped babysitting every single step most agents burn frontier tokens on tiny calls like "retry?" or "enough sources?". this one hands them to a model that does 10,000 decisions for ~$0.42 10 min and it runs on your machine: prompt → Claude → Jev picks the move → gate checks confidence → Claude executes → logged result step 1 → get your Jev API key on the TypeSafe site step 2 → clone the router from the GitHub below and install it in one line: python3 -m venv .venv && .venv/bin/pip install -r requirements.txt && cp config.example.yaml config.yaml step 3 → export the key before the first run: export TYPESAFE_API_KEY='YOUR_KEY' put the same line in ~/.zshrc or ~/.bashrc so a fresh terminal doesn't lose it step 4 → give Claude skill/jev-decision-layer.SKILL.md and point it at src/router.py + recipes/. on its own Jev only returns probabilities, the router is what turns them into actions Claude can run step 5 → test the full chain end to end, not just the Jev endpoint: .venv/bin/python -m src.cli '{"goal":"what is 2+2?","kind":"chat"}' the JSON you get back should have action, reason, mode, jev_used and confidence step 6 → start in mode: shadow. Claude works like before while every Jev call is logged next to what Claude actually did. after 20-50 real decisions keep only the question packs that matched step 7 → flip to mode: active with hard gates: 0.80+ acts on its own, 0.50-0.79 only suggests, under 0.50 comes back to you what changed for me: Claude stopped wasting tokens deciding whether to retry, reuse the cache or spin up a subagent. Jev makes those calls, Claude does the actual work, and anything it isn't sure about lands on my desk instead of in my outbox grab the exact stack from the repo, then read how every gate works in the full Jev architecture breakdown ↓

shawn

@shawnbuilds

Claude Mods are like custom Minecraft servers Claude Code Here's how to use Claude Code Mods + Jev to build your own custom memory harness for Claude Code. It runs in milliseconds, at sub-penny cost compared to subagents ⬇️ x.com/i/article/2106…

Ethan Walker

@ethan_walker_ai

🧵 What if the next big AI model isn’t designed to chat? TypeSafe AI is pushing a different direction with Jev, a model built to make structured decisions for software. No essays. No long answers. Just decisions that software can act on. Here’s why this matters 👇