Agent Loop Detector 🔁

Paste an autonomous agent's run (one step per line) and Jev grades each step — progress / repeating / stuck / escalate — then flags when the agent is looping so a supervisor can halt it before it burns tokens. Copy the flagged steps. Every decision is one real call to Jev, TypeSafe AI's System One model — a typed, calibrated decision (a choice, score, or yes/no probability) in ~100–300 ms, no chat, no hallucinations.

▶ Play live (no setup): https://jevtypesafeai.com/tools/agent-loop-detector

Run it locally

A single, dependency-free index.html — no build, no framework.

1. Open index.html in a browser.
2. Click ⚙︎ Jev API key, paste a key, Save.
3. Play.

Get a key

• jevtypesafeai.com — a key in a minute, no waitlist (endpoint is browser-ready).
• Or an official TypeSafe key — switch the endpoint in settings (the official API may block direct browser calls via CORS; use jevtypesafeai.com or a proxy).

Your key is stored only in your browser's localStorage and sent straight to the endpoint you pick.

How it works

Each step is one Jev choice call; a sustained run of non-progress trips a "loop detected — intervene" verdict. Wrap your agent loop in it and auto-halt on STUCK / ESCALATE.

const res = await fetch("https://jevtypesafeai.com/api/v1/decide", {
  method: "POST",
  headers: { "Content-Type": "application/json", Authorization: `Bearer ${JEV_KEY}` },
  body: JSON.stringify({ state, questions }),   // situation + typed questions
});
const { answers } = await res.json();           // typed decision + probabilities

The same primitive you'd drop into a real product. More: live page · all Jev tools · API docs.

---

Made with Jev · MIT licensed · PRs welcome.