Jev + Decision Models: Let Agents Decide Only on Change
Author shares combining Jev with decision models and adding a tinyfish monitor so the agent only decides when something has changed.
THE JEV GUIDE · CURATED EDITION
Launch news, explanations, demos, and community projects around TypeSafe Jev, organized by topic. Every entry links back to its X source.
Curated selection · Not exhaustive · Updated manually
01 / EXPLORE
0 means awaiting a verified post
Author shares combining Jev with decision models and adding a tinyfish monitor so the agent only decides when something has changed.
Sharing a coding-agent pattern: /ticket-triage asks Jev where to look, Cursor checks the repo, Jev flags test gaps after the diff, PHPUnit/Playwright verify behavior, and a human reviews before commit.
The open-source Jev assistant reads the screen via accessibility services, uses AI to assess intent and risk level, generates three reply suggestions, and supports one-tap fill without auto-sending.
Koimiao's team built a Stanford Town simulation with Jev. It's not a game demo—there is no player. Residents decide where to go, who to meet, and when to talk, while humans can only watch. Includes demo and GitHub code. Link
By having Jev act as a fast, intuitive gatekeeper before heavy LLM generation, the hybrid pipeline cuts token costs and latency while preserving high-quality memory extraction and knowledge graph construction.
Jev Engineering's Autopilot targets the gap where AI agent swarms execute well but don't know when to stop, as seen with agents running up to 4,000 steps.
A user used Jev to control an Android phone, autonomously opening the Uber app, entering the trip, and reaching the payment screen in 9 actions within about 21 seconds.
The author shows how Jev quickly finds the entry and accurately fills in external link info, and plans to switch the previous Submit Agent to JEV mode.
The post shows Luna alone takes ~39 calls, ~5 min, ~$0.037 per win, while Jev+Luna takes ~4 calls, ~1.7 min, ~$0.009 per win, highlighting Jev's efficiency and cost advantages.
Luna stays quiet until Jev repeats the same move three times without progress, then gives a short tip. Jev still makes every move; Luna only points.
JevTree expands multiple candidate futures and merges equivalent states to turn local intuitions into a long-horizon probability graph, enabling Jev to autonomously handle multi-step decisions.
JEV surpassed 18K stars in a week. jev-ultrafast splits every browser decision into operation, click target, and typed text, so one LLM request replaces five. It books Zurich to London on Google Flights in 7.1 seconds, while ChatGPT Operator takes 3 minutes at $200/month.