Skip to content

Jev directory

Jev plays Zork

An experiment where Jev plays Zork I: every turn Jericho supplies the valid actions and Jev answers one Choice of which command to type, with confidence bars, an anti-loop memory of the last 8 turns, a JSONL log and a French replay dashboard.
Stars
1
Language
Python
License
MIT
Reading time
3 min
Published by
Community
Author
resadan-dev
Last verified
Oct 2026

Highlights

  • Each turn asks three questions in one request: a Choice over the valid actions (criteria keys, at most 255), a noul danger check and an intent Choice.
  • The last 8 turns go into the state; an action already tried in the same world state has its probability halved on every retry.
  • Jev only judges: memory, anti-loop and the played command stay in code, and departures from Jev's opinion are flagged in terminal, log and replay.
  • Three parts: a terminal harness writing JSONL, replay/index.html with map and confidence curve, and video/render_video.py producing MP4 and GIF.
  • State is written in English because Jev reads it better; the game dashboard, replay player and videos are French.

Quickstart

bash
play install
# in PowerShell: .\play install

Watch out

Grounded in the truncated repo README and GitHub metadata reviewed on 2026-10-15; the README is cut off in the setup section, the project is a single-author MIT experiment with no benchmarks, Jericho must be compiled with gcc and make, and the Zork ROM is downloaded at setup because it is not versioned.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai