LLM World Engine Knowledge Base

LLM World Engine Knowledge Base

3 min read

Welcome to the LLM World Engine Knowledge Base

Comprehensive documentation extracted from 24 months of Discord discussions + ChatBotRPG source code analysis

This knowledge base synthesizes 12,109 Discord messages from the SillyTavern community’s “LLM World Engine” thread (January 2024 - December 2025) into production-ready documentation comprising 139 files and 63,000+ lines of content.

📚 Quick Navigation

Core Resources

Master Index - Complete navigation hub for all 139 documents

Main Topics

  1. Architecture & Design - Core philosophy and system design
  2. Prompt Engineering - Production-tested prompt templates
  3. RAG & Memory - Retrieval systems and context management
  4. World Generation - Procedural content generation
  5. State Management - Game state and persistence
  6. UI & Frontend - User interfaces and interactions
  7. Models & APIs - LLM integration patterns
  8. NDL Specification - Natural Description Language

🎯 Documentation Collections

Prompt Library (18 Templates)

Production-tested prompts with real performance metrics:

  • Narration (5): NDL translation, scene/action/dialogue/combat
  • Generation (3): Character, location, template creation
  • Constraint (4): Anti-hallucination, format enforcement, length limiting
  • Retrieval, Reasoning, System (6): HyDE queries, chain-of-thought, few-shot

Pattern Library (21 Patterns)

Battle-tested architectural patterns:

  • Architectural (4): Program-first, LLM pipeline, separation of concerns, event-driven
  • Integration (4): NDL bridge, state injection, API abstraction, multi-model routing
  • State (3): Three-tier persistence, scene boundaries, conditional saves
  • Generation (3): JIT generation, hierarchical cascade, template meta-generation
  • Control (4): Constraint prompting, chain-of-thought, few-shot, temperature switching

NDL Specification (27 Documents)

Complete Natural Description Language specification:

  • 7 Core Constructs: do(), search(), wait(), result(), sequencing, manner, intention
  • 4 Formal Specs: EBNF grammar, lexical elements, semantics, type system
  • 2 Integration Guides: LLM pipeline, parser implementation

ChatBotRPG Analysis (41 Documents)

Deep implementation analysis:

  • Discord-based analysis: Pattern implementation, production lessons
  • Code analysis: 15 extracted prompts, 100+ code references, 89% validation accuracy
  • Schemas: 10 core + 8 sub-schemas documented
  • Architecture: Multi-provider API integration

🔍 Key Insights

“LLMs as Narrators, Not Decision-Makers”

The fundamental philosophy: Game logic runs in deterministic code, NDL captures decisions, LLMs translate to natural language.

Production Metrics

  • 170-token sweet spot: Optimal prompt length
  • 40% cost savings: Structured state vs. verbose context
  • 95%+ anti-hallucination accuracy: Constraint-based prompting
  • 8B models viable: Gemma 2 9B sufficient for narration

Key Contributors

📊 Project Statistics

  • 139 total files: Comprehensive knowledge base
  • 63,563 lines: Production-ready documentation
  • 1,000+ code examples: Python with type hints
  • 41+ Mermaid diagrams: Architecture visualizations
  • 24 months synthesized: Jan 2024 - Dec 2025
  • 15+ contributors: Community-driven knowledge

🚀 Getting Started

New to LLM game engines? Start here:

  1. Read Architecture & Design to understand the core philosophy
  2. Explore NDL to Narrative - the foundational prompt
  3. Review Program-First Architecture pattern
  4. See ChatBotRPG for a real implementation

📖 How to Use This Knowledge Base

  • Browse by topic: Use the topic indexes above
  • Search: Use the search box (top left) to find specific content
  • Follow links: Click wiki-links to navigate between related documents
  • View graph: See the knowledge graph (right sidebar) for visual connections
  • Check backlinks: Each page shows what other pages reference it

🤝 About This Project

This knowledge base was created using automated AI agents that:

  1. Analyzed 12,109 Discord messages over a two year period (Jan 2024 - Dec 2025)
  2. Extracted prompts, patterns, diagrams, and other technical documentation
  3. Validated discussions and claims against ChatBotRPG source code (89% accuracy)
  4. Generated Obsidian-compatible markdown with proper wiki-links

Source: SillyTavern Discord - #llm-tech-chat channel, llm-world-engine thread

Last updated: January 2026 | Built with Quartz v4

Graph View