ZorkGPT is an AI agent that plays the classic interactive fiction game Zork I using a Large Language Model. It navigates a text-only world, decides what to type next, and learns the map and the rules of the place purely by playing — no walkthrough and no game-specific hints are given to it.
What is Zork?
Zork is a legendary text adventure game first released in 1977. Players explore a sprawling underground world by typing natural-language commands, solving puzzles, collecting treasures, and navigating hundreds of locations described entirely in text.
What am I looking at?
This viewer shows live data from a running ZorkGPT session:
Objectives: Goals the agent has discovered on its own, with the exact rule that has to hold for each one to count as done
Game Log: Every command the agent sends, its stated reason for sending it, and the game's response
Map: Rooms discovered and the passages between them
Untried Directions: Exits from the current room the agent has never tried
Score / High Score Progression: How the score has moved within this run, and the best score across every run this viewer has archived
How does the agent decide what to do?
Each turn the agent reads the room description and its own recent history, then commits to exactly one command. The command and the reason given for it are both recorded before the game engine runs it, so the log you see here is the agent's actual reasoning, not a summary written after the fact. The room, score, and inventory shown throughout this viewer come straight from the game engine — never from the agent's own account of what happened.
What do the log colors mean?
Yellow — The reason the agent gave for its command (click to expand)
Light green — The command sent to the game
Green — The game's response
What does the Objectives panel show?
Objectives are goals the agent identifies for itself during play, not a fixed list handed to it in advance. An open objective shows the exact condition that will satisfy it; a completed one shows the evidence — the game state that was true at the moment it was judged satisfied.
Why isn't the data updating?
The viewer polls every 3 seconds. A red dot means one of:
No ZorkGPT session is currently running
The last recorded turn is more than 10 minutes old
A network connectivity issue
Can I interact with the game?
No — this is a read-only viewer. The agent plays autonomously.