Zing Forum

Reading

Unarmored: When Large Language Models Become Narrative Engines for Medieval RPGs

This article introduces Unarmored, a text adventure game centered on large language models (LLMs) as its narrative core, exploring AI-driven dynamic storytelling, character state tracking, skill progression systems, and methods for building historical immersion.

大型语言模型文字冒险游戏RPGAI 叙事中世纪游戏设计动态叙事
Published 2026-05-17 23:46Recent activity 2026-05-17 23:52Estimated read 6 min
Unarmored: When Large Language Models Become Narrative Engines for Medieval RPGs
1

Section 01

Unarmored: Introduction to an LLM-Powered Dynamic Narrative Engine for Medieval RPGs

Unarmored is a medieval text role-playing game (RPG) centered on large language models (LLMs) as its narrative core. Set in the Silesia region in 1325, players take on the role of Werner, the son of a tanner, interacting with the AI by inputting actions to experience dynamically generated, historical novel-style storytelling. The game integrates classic RPG mechanics (character progression, skill systems, economic systems) with AI technology, emphasizing historical immersion and realistic consequences, blurring the line between games and interactive literature.

2

Section 02

Game Background and Construction of Historical Immersion

Unarmored is set in Silesia in 1325 during the fragmentation of the Piast dynasty. The protagonist Werner inherits his father's tanning workshop, debts, and orders from the lord. The game deliberately avoids fantasy elements, using material details like leather, mud, and debts to build an authentic medieval texture. Locations include the tanning workshop, the Grey Goose Tavern, the priest's residence, etc., with details such as bulletin boards enhancing scene credibility. The initial debt and oath (to make leather armor for the lord's steward) are the core driving forces.

3

Section 03

Technical Implementation of the AI Narrative Engine

The game's narrative is based on LLMs. Each turn, it receives the complete game state (character attributes, skills, items, etc.) as context. When generating content, it implicitly references data but does not directly mention numerical values (e.g., a tanning skill of 12 is described as "clumsy", while 45 is described as "skilled and precise"). To solve the problem of long dialogue context, an "Archivist" mechanism is introduced: after every 12 messages, a lightweight model summarizes the old dialogue into an archive, retaining up to 4 summaries to balance narrative coherence and performance.

4

Section 04

Character Progression and System Design

Character skills are divided into three categories: Combat (swordsmanship, wrestling, etc.), Knowledge (reading, herbalism, etc.), and Craftsmanship (tanning, blacksmithing, etc.), ranging from 0 to 100, and are improved through actual use. The crafting system uses opportunistic progression: success grants experience and items, while failure consumes materials but still gives a small amount of experience. Social tracking covers 6 factions (such as Hartburg Village, the Church) and relationships with 6 residents, reflected through changes in NPC behavior without directly displaying numerical values.

5

Section 05

Unique Gameplay and Experience Enhancement

The workshop system includes 5 tabs such as tanning, blacksmithing, bowmaking, etc., requiring corresponding skills and materials; equipment is divided into 10 parts, which only take effect when worn. The visualization function converts the narrative into Bruegel-style images (supporting multiple AI backends); environmental sound effects automatically switch between combat/ambient music, and TTS reads the narrative. Saves are in JSON format, supporting manual/automatic saving to facilitate exploring multiple paths.

6

Section 06

The Value and Significance of Unarmored

Unarmored represents a new direction for AI-driven games: using LLMs as real-time narrative engines to create dynamic story experiences, blurring the line between games and interactive literature. For players, it provides a slow-paced, contemplative experience that combines the texture of classic text adventures with the infinite possibilities of modern AI; for developers, it is a case study in AI game design, demonstrating the balance between state management, prompt engineering, and UX, with a modular architecture that facilitates expanding new content.