RealmForge

RealmForge

Procedural content generation for games using LLMs. Generates scenes, characters, NPCs, and quests through a FastAPI server, with optional Three.js visualization.

Architecture

Directory	Purpose
src/api/	FastAPI server with routers for world, NPC, narrative, difficulty, visualization
src/models/	Pydantic models for world, NPC, narrative, difficulty, player, visualization
src/services/	Business logic: NPC generation, narrative, difficulty scaling, optimization
src/config/	Settings (env-based configuration)

What It Generates

Content type	Description
Scenes	3D environments (forests, towns, caves, dungeons)
Characters	Attribute sets and appearance descriptions
NPCs	Personalities, behaviors, backstories
Quests	Objectives, narrative text, rewards

Generation is done via LLM calls orchestrated through an AFLOW-style workflow pipeline.

Requirements

• Python 3.11+
• API keys for your LLM provider (configured in .env)

Setup

git clone https://github.com/jmanhype/RealmForge.git
cd RealmForge
pip install -e .
cp .env.example .env  # add API keys
python -m src.api.main

Server runs at `localhost:8001`. API docs at `localhost:8001/docs`. Visualization at `localhost:8001/visualizer`.

Status

Prototype. The generation endpoints work but output quality depends entirely on the LLM and prompts used. The Three.js visualizer renders basic scenes. The roadmap lists game engine integrations (Unity, Unreal) and avatar format support (VRM) -- none of that exists yet. No tests are included.

Note: the repository contains `__pycache__` directories and `.pyc` files that should have been gitignored.

License

MIT