Static Vs. Agentic Game Master AI for Facilitating Solo Role-Playing Experiences

TL;DR

This work tackles the challenge of delivering D&D-like solo role-playing through AI by developing ChatRPG, a text-based Game Master (GM) system. It contrasts two designs: a baseline v1 relying on prompt engineering and a ground-up v2 using a multi-agent ReAct framework with distinct Narrator and Archivist agents to separate storytelling from memory management. Empirical evaluation shows that v2 yields higher immersion, perceived intelligence, and user engagement, validating the benefits of agentic AI and structured tool use for dynamic, coherent narratives. The study underscores the potential of agentic LLMs to augment solo IF experiences, discusses design trade-offs and API limitations, and provides open-source code to encourage replication and extension.

Abstract

This paper presents a game master AI for single-player role-playing games. The AI is designed to deliver interactive text-based narratives and experiences typically associated with multiplayer tabletop games like Dungeons & Dragons. We report on the design process and the series of experiments to improve the functionality and experience design, resulting in two functional versions of the system. While v1 of our system uses simplified prompt engineering, v2 leverages a multi-agent architecture and the ReAct framework to include reasoning and action. A comparative evaluation demonstrates that v2 as an agentic system maintains play while significantly improving modularity and game experience, including immersion and curiosity. Our findings contribute to the evolution of AI-driven interactive fiction, highlighting new avenues for enhancing solo role-playing experiences.

Figures (9)

• Figure 1: An example of a ReAct reasoning trajectory.
• Figure 2: Screenshots of the ChatRPG game: a) Landing page of the game. b) Example of a campaign and the text-based conversational user interface of the game.
• Figure 3: Game interaction flow diagram showing how in v1 user input is handled by the system to make calls to the LLM and present updates to the UI.
• Figure 4: Game interaction flow diagram showing how, in v2, user input is handled by the Narrator and Archivist agents to make tool calls and prompts to the LLM and present updates to the UI.
• Figure 5: This figure illustrates the architecture of ChatRPG v2, which integrates user input, AI reasoning, and a dynamic world state. The system starts with the Front-End's Text-Based Interface, where players input their actions. These inputs are processed by the Back-End's Game Input Handler and passed to the Narrator Agent, which uses the ReAct framework to generate a narrative through decision-making. Reasoning and resolutions are handled via the OpenAI API, and Tools are employed when specific actions are required. The Archivist Agent ensures changes are recorded in the Campaign World State, which is stored persistently in the Database using Entity Framework. The closed loop allows for continuous gameplay driven by player input and AI responses.