Project Infinity: A Dynamic, Text-Based RPG World Engine

Project Infinity is a sophisticated, procedural world-generation engine and AI agent architecture. It demonstrates a novel solution to several critical challenges in modern AI, including state management, factual consistency, and the creation of highly efficient, specialized agents. The latest version introduces a radically improved agent protocol that enables more dynamic, emergent storytelling and achieves a new level of LLM-agnostic portability.

A Case Study in Next-Generation AI Architecture

This project serves as a proof-of-concept for building highly capable, consistent, and secure AI agents. By integrating a procedural generation engine with a knowledge-grounded Large Language Model (LLM), Project Infinity successfully overcomes several critical challenges in the field.

Key Innovations

• Knowledge-Grounded Generative System (Graph RAG): At its core, Project Infinity utilizes a Graph RAG architecture. A "World Forge" engine first generates a comprehensive knowledge graph (The Key) that serves as a "single source of truth" for the AI. This graph is not just a list of entities, but a deeply interconnected world model of lore, politics, and geography. Grounding the agent in this graph solves the core problem of model hallucination.

• Codified Agent Protocol: The project's primary innovation is its method for agent specialization. The GameMaster.md file (The Lock) is not a natural language prompt, but a highly structured, token-efficient protocol. Written as a YAML-based schema, it defines the agent's core logic, operational states, and behavioral directives in a format optimized for LLM-to-LLM communication. Crucially, the protocol now includes priming meta-instructions, making it robustly compatible across different foundational models (including Gemini, ChatGPT, and Mistral), ensuring the agent behaves consistently in any environment.

• Proprietary Narrative Engine (L.I.C. Matrix): Beyond simple factual retrieval, the agent's storytelling is governed by the L.I.C. (Logic, Imagination, Coincidence) Matrix. This proprietary framework acts as an "imagination driver," guiding the AI to weave facts from the knowledge graph with emergent story elements in a way that feels meaningful, creative, and alive.

Broader Implications

While demonstrated within a complex gaming simulation, the architecture of Project Infinity serves as a powerful blueprint for a new class of enterprise-grade AI agents. The project's success in achieving stateful consistency and intrinsic security via its codified protocol presents a viable path forward for developing specialized AI that is not only highly capable but also reliable and safe for critical applications.

---

Technology Stack

• Backend: Python 3
• Data Validation: Pydantic
• Configuration: PyYAML
• Procedural Generation: NumPy, noise

The "Lock & Key" System

The engine's core design principle is the separation of the agent's rules from the world's data.

• The Lock (GameMaster.md): This file is the Codified Agent Protocol (3.7 KiB). It is a YAML-based schema that instructs a general LLM on how to interpret world data, manage game mechanics, and execute its core logic. The protocol is LLM-agnostic, ensuring consistent agent behavior across different foundational models.

• The Key (output/<character_name>_weave.wwf): This is the Knowledge Graph. It is a pre-generated world-state file that contains the core, static data of a unique world. The latest version uses a schema-driven, positional array format that reduces the file size from 27.7 KiB to a final, hyper-efficient 10.3 KiB.

Getting Started

1. Prerequisites

• Python 3.8+
• git

2. Installation

# Clone the repository
git clone https://github.com/electronistu/Project_Infinity
cd project_infinity

# Create and activate a Python virtual environment
python3 -m venv venv
source venv/bin/activate

# Install the required dependencies
pip install -r requirements.txt

3. World Generation (Optional)

To forge your own unique world, run the main script:

python3 main.py

This will launch the interactive character creator. Follow the prompts to build your character, after which the Forge will generate your world. The output will be saved as a new .wwf file in the output/ directory, named after your character.

For development, you can bypass the interactive prompts using the --debug flag:

python3 main.py --debug

How to Play

This project includes a pre-generated world file, output/electronistu_weave.wwf, so you can start playing immediately.

Compatible Platforms

The protocol is designed to be LLM-agnostic and has been successfully tested on the following platforms. For best results, use the latest available models and set the Temperature to 0 for maximum consistency.

• Google: Gemini 2.5 Pro (via AI Studio, Gemini CLI, etc.)
• OpenAI: ChatGPT-5
• Mistral AI: chat.mistral.ai

The "Lock & Key" Process:

1. Load the "Lock": Start your session by providing the contents of the GameMaster.md file to your chosen AI platform.

2. Await Confirmation: The AI should respond with the words: Awaiting Key...

3. Provide the "Key": Paste the entire contents of the generated .wwf file (e.g. output/electronistu_weave.wwf).

4. Begin Your Adventure: The Game Master will parse the world and begin your unique, text-based adventure.

Emergent Agent Personas

A fascinating outcome of this project is observing the distinct "personalities" that emerge when the same GameMaster.md protocol is executed by different foundational models. While the core rules and logic remain identical, the flavor of the Game Master changes, revealing the unique architectural biases of each LLM.

• Gemini as "The Cinematic Narrator": Gemini tends to produce a highly immersive, story-focused experience. Its output is often cinematic, with descriptive prose that sets a rich scene and immediately draws the player into a narrative, much like the opening of a film.

• ChatGPT as "The Interactive Guide": ChatGPT often adopts the role of a classic Game Master. It presents the world in a slightly more gamified manner, clearly outlining choices (often with numbered lists) and explicitly referencing game concepts, creating an experience reminiscent of a classic gamebook.

• Mistral as "The World Simulator": Mistral acts like a data-rich world simulator. Its output is incredibly structured, often presenting the player with a detailed dashboard of the current world state, including emergent quests, notable NPCs with stats, and environmental details. This empowers the player with a wealth of information, encouraging tactical and strategic decision-making.

This demonstrates that even with a rigid, codified protocol, the underlying model's "imagination" still shapes the final experience, making the choice of LLM a creative decision in itself.