This project provides a proof-of-concept GPT-based agent operating within a simple, locally-run RPG-like environment. It's designed for developers and researchers interested in exploring LLM agent behavior and interaction within simulated worlds. The core benefit is a functional demonstration of an AI agent making decisions and interacting with a game environment using the OpenAI API.

How It Works

The system comprises two main components: an RPG environment and an AI agent. The environment, built with React, Tiled, Phaser, and the Grid Engine Plugin, features a character, impassable tiles, and plantable layers. The agent, a Node.js application, connects to the OpenAI API (specifically gpt-3.5-turbo) via websockets to receive environmental state and available actions, then outputs decisions. This architecture allows the LLM to drive agent behavior in a visually represented, interactive setting.

Quick Start & Requirements

• Install dependencies: npm install in the gptrpg directory.
• Run: npm start in the root directory.
• Prerequisites: Node.js (tested with 16.19.0), OpenAI API key (configured in agent/env.json).
• Access: Frontend available at http://localhost:3000.

Highlighted Details

• Demonstrates LLM agent interaction within a simulated RPG environment.
• Utilizes OpenAI API (gpt-3.5-turbo) for agent decision-making.
• Environment built with React, Tiled, Phaser, and Grid Engine Plugin.
• Agent communicates with the frontend via websockets.

Maintenance & Community

No specific community channels, roadmap, or contributor information is provided in the README.

Licensing & Compatibility

The license is not specified in the README.

Limitations & Caveats

The project is explicitly a proof of concept and currently only supports gpt-3.5-turbo. Upcoming features like multi-agent support, expanded actions, agent memory, and goals are not yet implemented. The README notes that the plant layer is not currently in use by the agent.