RAG System

The Retrieval-Augmented Generation (RAG) system provides accurate, up-to-date Minecraft information by pulling from a knowledge base.

What is RAG?

RAG combines:

  • Retrieval: Finding relevant information from a knowledge base
  • Generation: Using that information to create accurate responses

This ensures the ServerAI provides factual, current Minecraft information. (RAG is currently wired to ServerAI only.)

Requirements

LM Studio Setup

RAG requires LM Studio with a specific embedding model:

  1. Download LM Studio: Get it from lmstudio.ai
  2. Install Embedding Model: Search for and download “nomic-ai/nomic-embed-text-v1.5-GGUF”
  3. Alternative Source: Download from HuggingFace
  4. Load Model: Start LM Studio and load the embedding model
  5. Start Server: Ensure LM Studio server is running

Configuration

Enable RAG in config.yml:

server-ai:
  rag:
    enabled: true
    dataPath: "ragData"
    embeddingModel: "nomic-ai/nomic-embed-text-v1.5"
    lmStudioUrl: "http://localhost:1234"
    maxContextLength: 2000
    topK: 5
    maxHops: 1
    similarityThreshold: 0.5
    maxTotalDocs: 15
    maxContextDocs: 5
    snippetWindow: 600
    fallbackPrefixLen: 800

How RAG Works

Knowledge Base Creation

  1. Data Collection: Gather Minecraft information (wiki pages, guides, etc.)
  2. Text Processing: Split content into manageable chunks
  3. Embedding Generation: Convert text to numerical vectors
  4. Storage: Save embeddings in the ragData directory

Query Processing

  1. Question Analysis: User asks a question
  2. Similarity Search: Find relevant knowledge chunks
  3. Context Assembly: Combine relevant information
  4. Response Generation: AI uses context to answer accurately

Configuration Options

Basic Settings

  • enabled: Turn RAG on/off
  • dataPath: Directory for knowledge base storage
  • lmStudioUrl: LM Studio server URL

Embedding Model

  • embeddingModel: Must be “nomic-ai/nomic-embed-text-v1.5”
  • Requires LM Studio to host the model

Search Parameters

  • topK: Initial documents to retrieve (default: 5)
  • maxHops: Graph expansion depth (default: 1)
  • similarityThreshold: Minimum similarity for expansion (0.0-1.0)
  • maxTotalDocs: Maximum documents after expansion (default: 15)
  • maxContextDocs: Documents included in final context (default: 5)

Content Processing

  • snippetWindow: Characters around matched phrases (default: 600)
  • fallbackPrefixLen: Characters to include if no match (default: 800)
  • maxContextLength: Maximum context length (default: 2000)

Setting Up Knowledge Base

Downloading Pre-built Knowledge Base

A pre-built knowledge base is available for download:

  1. Download the ZIP: Get ragData.zip directly from the repository at ragData/ragData.zip.
  2. Extract: Unzip the file into your plugins/Chatr/ folder.
  3. Verify: Ensure the ragData folder contains the required files (see File Structure below).

This provides a ready-to-use knowledge base with current Minecraft information.

Manual Data Addition (Advanced)

For custom knowledge bases, you can manually populate the data. This requires Python and LM Studio setup. Future versions may include automated data collection.

Data Format

Knowledge base expects text documents containing Minecraft information:

  • Crafting recipes
  • Block/item information
  • Game mechanics
  • Update changelogs
  • Community guides

File Structure 

plugins/Chatr/ragData/
├── docs.json           # Document text and metadata
├── embeddings.npy      # Numpy array of document embeddings
├── faiss_index.index   # FAISS similarity search index
├── graph.json          # Document relationship graph
└── *.py                # Python scripts for building/reranking (optional)

Performance Considerations

Resource Usage

  • Embedding generation requires GPU/CPU resources
  • Storage space for knowledge base
  • Memory for context processing

API Costs

  • RAG reduces API calls by providing accurate information
  • Local embedding model (no cloud costs)
  • Only generation uses configured AI provider

Response Quality

  • More accurate Minecraft information
  • Reduced hallucinations
  • Context-aware responses

Troubleshooting

Model Loading Issues

  • Ensure nomic-ai model is downloaded and loaded in LM Studio
  • Check LM Studio server is running on correct port
  • Verify model name matches exactly

No Context Retrieved

  • Check knowledge base has been populated
  • Verify similarity threshold isn’t too high
  • Test with simpler queries first

Poor Response Quality

  • Ensure knowledge base contains relevant information
  • Adjust maxContextDocs for more/less context
  • Check AI provider configuration

Performance Issues

  • Lower topK and maxTotalDocs for faster searches
  • Increase similarityThreshold to reduce noise
  • Monitor LM Studio resource usage

Best Practices

Knowledge Base Management

  • Regularly update with new Minecraft information
  • Include diverse sources (official wiki, community guides)
  • Test knowledge base with common questions

Configuration Tuning

  • Start with default settings
  • Adjust based on server performance
  • Monitor response quality and speed

Content Quality

  • Use accurate, up-to-date information
  • Include examples and edge cases
  • Organize content by topic

Future Enhancements

Planned improvements:

  • Automated knowledge base population
  • Web scraping for current information
  • Multiple embedding model support
  • Knowledge base management tools

See Also