Guide: building a self-running agent loop with Opus 5.5 and Jev
Opus 5.5 does the thinking while Jev returns three structured outputs (Choice, Score, Noul) to decide the next action, whether output passes, and whether the goal is done, with first calls in about 10 minutes.
this is f*cking gold. Opus 5.5 + Jev: how to build your first AI agent that runs itself. Everything you need to be ahead of most people. Opus 5.5 does the thinking. Jev makes every decision in the loop: what your agent does next, whether its output is good enough, and when the goal is done. You don't have to watch it. Jev is a typed decision model built for agent loops. It reads your system state and returns one of three structured answers: Choice - which agent or action runs next? Score - is this output good enough to pass? Noul - is the goal complete? the architecture: Opus 5.5 thinks. Jev decides. Agent executes. About 10 minutes to your first call. Hosted on Vercel and Cloudflare via Typesafe, no weights to download. Access is through a waitlist, and people report getting in within 24 hours. The guide in the image covers the full setup: architecture, access, install, and your first decision call. Full breakdown on agents, loops, and costs below.
This post quotes another post: x.com/loofyb0i/status/2102452178384273431