ServerAI

The ServerAI is a server-wide AI assistant that monitors chat and helps players proactively.

Overview

ServerAI provides:

  • Chat monitoring and automatic responses
  • Proactive help and conversation joining
  • Server-wide assistance without specific NPC interactions
  • Configurable personality and behavior

Enabling ServerAI

Set in config.yml:

server-ai:
  enabled: true
  name: "Heimdall"

Configuration Options

Basic Settings 

server-ai:
  enabled: true
  name: "Heimdall"                    # Assistant name
  display:
    show-prefix: true                 # Show [AI] prefix
    prefix-text: "AI"                 # Prefix text
    prefix-color: "&6"                # Gold color
    name-color: "&a"                  # Green color

Chat Monitoring 

server-ai:
  chat-scan-interval-seconds: 60      # Check chat every 60 seconds
  conversation-join-chance: 0.1       # 10% chance to join conversations
  max-chat-history: 25                # Remember last 25 messages
  max-conversation-memory: 10         # Messages per player conversation

AI Settings 

server-ai:
  server-url: "http://localhost:1234"  # Independent of global ai.endpoint
  model: "local-model"                 # Independent of global ai.model
  max-tokens: 2000                      # Response length limit

How ServerAI Works

Chat Monitoring

  • Scans recent chat messages periodically
  • Looks for opportunities to help or join conversations
  • Responds to direct mentions of its name

Conversation Joining

  • Randomly joins ongoing conversations based on conversation-join-chance
  • Provides helpful information or mediates conflicts
  • Only responds when it can add value

Direct Interaction

  • Players can mention the AI by name in chat
  • Example: “Heimdall, how do I make a diamond pickaxe?”
  • AI responds naturally in chat

System Prompt

Customize the AI’s personality and behavior:

server-ai:
  system-prompt-source: "inline"  # "inline" or "file"
  system-prompt: |
    You are {name}, the Server AI assistant for this Minecraft server.

    RESPONSE RULES:
    - Respond ONLY with your message. No analysis, no thinking, no prefixes.
    - Do NOT include location, coordinates, biome, weather, or time information unless asked.
    - Keep responses focused on the player's query.
    - Just respond naturally as if you're chatting.
    - Use plain text only. No markdown formatting.

    Your personality:
    - Wise and calm, like an all-seeing guardian
    - Helpful and knowledgeable about Minecraft
    - Friendly but professional

Context Variables

ServerAI supports the same context variables as NPCs:

  • {name} - AI’s name
  • {server_time} - Current server time
  • {player_name} - Mentioning player’s name
  • {weather} - Current weather
  • And more…

Limitations

AI Provider Compatibility

Important: ServerAI only works with OpenAI-compatible HTTP APIs:

  • ✅ LM Studio (default)
  • ✅ Ollama
  • ✅ Groq
  • ✅ Mistral
  • ✅ NVIDIA
  • ✅ OpenRouter
  • ❌ Gemini (not compatible)

For Gemini support, use individual NPCs instead.

File-based Prompts

You can load prompts from a file:

server-ai:
  system-prompt-source: "file"

Then create plugins/Chatr/serverAI/system-prompt.txt with your custom prompt.

Best Practices

Personality Design

  • Keep the personality consistent and helpful
  • Use context variables for dynamic responses
  • Focus on being a positive community presence

Performance Tuning

  • Adjust chat-scan-interval-seconds based on server activity
  • Lower conversation-join-chance for quieter servers
  • Monitor API usage with /chatr stats

Community Management

  • Use ServerAI to welcome new players
  • Have it provide general server information
  • Configure it to help with common questions

Troubleshooting

ServerAI Not Responding

  • Check that server-ai.enabled is true
  • Verify AI provider is configured and working
  • Ensure using OpenAI-compatible API (not Gemini)

Not Joining Conversations

  • Check conversation-join-chance is above 0
  • Verify chat-scan-interval-seconds is reasonable
  • Look for console errors during chat scanning

Wrong Personality

  • Check system prompt for correct formatting
  • Ensure context variables are properly used
  • Test with direct mentions first

See Also