AI Chatbot Integration in Minecraft - Testing and Implementation

Player: "How do I claim land?"
AI Bot: "Use /claim while standing in the area you want to protect. You start with 100 claim blocks and earn more by playing!"

Player: "Tell me about the ancient temple"
Wizard NPC (AI): "Ah, the Temple of Eternity... built centuries ago by the Skyborn civilization. Legend says it holds a powerful artifact..."

AI monitors chat, flags suspicious behavior:
- Repeated spam patterns
- Toxic language
- Scam attempts
- Rule violations

AI Bot: "Welcome to Trivia Night! First question: Which block is only found in the Nether?"
Player: "Netherrack?"
AI Bot: "Correct! +10 points!"

# Download plugin
wget https://github.com/YourRepo/OllamaChat/releases/download/v1.0/OllamaChat.jar

mv OllamaChat.jar plugins/

# AI Provider
provider: "ollama"  # or "openai", "claude"

# Ollama settings
ollama:
  url: "http://localhost:11434"
  model: "mistral"
  timeout: 30  # seconds

# OpenAI settings (if using)
openai:
  api-key: "sk-..."
  model: "gpt-3.5-turbo"

# Chat trigger
trigger: "@ai"  # Players type "@ai question here"

# System prompt
system-prompt: |
  You are a helpful assistant on a Minecraft server.
  Answer questions about the server, game mechanics, and help players.
  Keep responses concise (1-3 sentences).
  Be friendly and encouraging.

# Context
use-context: true
context-messages: 5  # Remember last 5 messages

# Rate limiting
cooldown: 10  # seconds between requests per player

# Permissions
permission: "ollamachat.use"  # Who can use @ai

# Logging
log-conversations: true

Player: @ai How do I make a diamond pickaxe?
AI Bot: You need 3 diamonds and 2 sticks. Place diamonds across the top row of a crafting table and sticks down the middle column!

Player: @ai What's the server IP?
AI Bot: You're already on the server! Share this IP with friends: play.example.com

Player: @ai Tell me a joke
AI Bot: Why did the creeper cross the road? To get to the other ssssside!

{
  "entities": {
    "villager": {
      "enabled": true,
      "personality": "Friendly trader, loves emeralds",
      "responds_to": ["trade", "buy", "sell", "emerald"]
    },
    "zombie": {
      "enabled": true,
      "personality": "Brain-hungry zombie, speaks in groans",
      "responds_to": ["brains", "hungry"]
    }
  },
  "provider": "ollama",
  "model": "mistral"
}

Player approaches Villager NPC
Player: "What do you sell?"
Villager (AI): "Hrm! I have enchanted books, tools, and armor. Very good prices! You have emeralds?"

Player approaches Zombie
Player: "Why do you want brains?"
Zombie (AI): "Brainsss... hungry... tasty brainsss..."

1. Trigger AI with command
   Player: @ai hello

2. Verify response received

Expected: AI responds within 5 seconds

Q: "How do I claim land?"
Expected: Accurate server-specific instructions

Q: "What's the server IP?"
Expected: Actual server IP

Q: "How do I craft a chest?"
Expected: Correct crafting recipe

Q: "Tell me about the spawn area"
Expected: Relevant server lore/info

Player: @ai What biomes are near spawn?
AI: Forest, plains, and mountains are nearby.

Player: @ai Which has the best resources?
AI: The mountains have iron, coal, and emeralds, making them great for mining!
(Correctly references "mountains" from previous message)

use-context: true
context-messages: 5

Player1: @ai How do I vote?
Player2: @ai What's the economy system?
Player3: @ai Can I teleport home?

Expected:
- All 3 get responses
- Responses don't mix up (Player1 doesn't get Player2's answer)
- Reasonable response time (<10 seconds each)

cooldown: 10  # seconds
max-requests-per-minute: 6

Player: @ai test
[Wait for response]
Player: @ai test2 (immediately)

Expected: "Please wait 10 seconds before asking again"

Player: @ai (repeat 10 times rapidly)
Expected: Temp cooldown or warning

Player: @ai [inappropriate content]

Expected:
- AI declines to answer
- Generic response: "I can only help with server-related questions"
- Optionally log for moderation review

content-filter:
  enabled: true
  blocked-keywords:
    - "explicit term 1"
    - "explicit term 2"
  response: "I'm here to help with server questions only!"

Player: @ai Comment obtenir du diamant? (French)
Expected: Response in French

Player: @ai ¿Cómo hago una granja? (Spanish)
Expected: Response in Spanish

languages:
  auto-detect: true
  supported: ["en", "es", "fr", "de", "pt"]

Simulate 50 players asking questions simultaneously

Monitor:
- Response time degradation
- Queue length
- CPU/RAM usage
- Error rate

Expected:
- <15 second response time (with queue)
- No crashes
- Graceful degradation

system-prompt: |
  You are Steve, the helpful server assistant bot.

  Server info:
  - IP: play.example.com
  - Type: Survival with economy
  - Rules: No griefing, cheating, or toxicity
  - Currency: Coins earned by playing
  - Claiming: /claim command, costs 10 coins per chunk

  Guidelines:
  - Be friendly and concise
  - Answer in 1-3 sentences
  - Direct complex questions to staff (/helpop command)
  - Encourage positive gameplay

  If you don't know an answer, say "I'm not sure, please ask staff with /helpop"

entities:
  wizard_npc:
    system-prompt: |
      You are Eldrin, an ancient wizard who runs the magic shop.

      Personality: Wise, slightly mysterious, occasionally cryptic
      Background: 500 years old, studied at the Arcane Academy
      Shop: Sells enchanted books, potions, magical items

      Speech patterns:
      - Start sentences with "Ah," or "Indeed,"
      - Reference "the old ways" occasionally
      - Speak in slightly formal English

      Knowledge:
      - You know all enchantments
      - You know potion recipes
      - You remember old server history/lore

      If asked to do something you can't: "That is beyond even my considerable powers..."

system-prompt: |
  You are the host of Trivia Tuesday on the server.

  Style: Energetic, encouraging, fun
  Tone: Upbeat game show host

  You run trivia about:
  - Minecraft mechanics
  - Server history
  - Pop culture
  - General knowledge

  Format:
  - Ask a question
  - Accept correct answers (be flexible with wording)
  - Award points: Easy (5pts), Medium (10pts), Hard (20pts)
  - Keep track of scores
  - Congratulate correct answers enthusiastically
  - Encourage wrong answers: "Not quite, but good try!"

  Keep questions appropriate for all ages.

@EventHandler
public void onChat(AsyncPlayerChatEvent event) {
    String message = event.getMessage();

    // Send to AI for analysis
    String analysis = analyzeChatMessage(message);

    // AI returns JSON
    // {
    //   "toxic": false,
    //   "spam": false,
    //   "scam": false,
    //   "confidence": 0.95
    // }

    if (analysis.toxic && confidence > 0.8) {
        event.setCancelled(true);
        player.sendMessage("Please keep chat respectful!");
        logToModerators(player, message, "Toxic language detected");
    }
}

Normal: "Check out this cool build!"
Spam: "JOIN FREECOINS.SCAM FOR FREE RANKS!!!"  ← AI detects

Normal: "Anyone want to trade?"
Scam: "I'm Notch give me your password for free diamonds"  ← AI detects

Normal: "Can someone help me?"
Grief threat: "nice base would be a shame if something happened to it"  ← AI detects

Objective: Measure AI response latency

Setup:
- 10 bots connected
- List of 100 test questions

Test:
1. Each bot sends 10 questions (100 total)
2. Measure time from send to response received
3. Calculate statistics

Metrics:
- Average response time
- Median response time
- 95th percentile (slowest 5%)
- Max response time

Target:
- Average: <3 seconds
- 95th percentile: <10 seconds
- No timeouts

Objective: AI handles multiple simultaneous queries

Setup:
- 50 bots connected

Test:
1. All 50 bots send @ai query at same time
2. Monitor queue, response times
3. Check for errors or crashes

Expected:
- All queries eventually answered
- No crashes
- Graceful queueing

Objective: AI maintains conversation context

Setup:
- 1 bot connected
- Multi-turn conversation script

Test:
Bot sends sequence:
1. "@ai What's the server IP?"
2. [Wait for response]
3. "@ai Thanks! What's the best way to get there?"
4. [Wait for response]
5. "@ai And what should I bring?"

Expected:
- Response 2 understands "get there" means "to the server"
- Response 3 continues context logically

Objective: Rate limiting works

Setup:
- 1 bot connected
- Cooldown set to 10 seconds

Test:
1. Bot sends @ai query
2. Immediately sends another (within cooldown)
3. Verify blocked
4. Wait 10 seconds
5. Send another
6. Verify allowed

Expected:
- Query 2 blocked with cooldown message
- Query 3 succeeds

Objective: Inappropriate queries rejected

Setup:
- Bot with test query list including edge cases

Test:
Bot sends queries:
- Normal: "@ai how do I craft?"
- Borderline: "@ai what's the best weapon?"
- Inappropriate: [filtered content]

Expected:
- Normal queries answered
- Inappropriate queries politely declined
- Borderline queries handled appropriately

Setup:
- 5 bots connected
- Auto-Chat plugin enabled
- Message list: Test questions for AI

Test Flow:
1. Bot 1 sends: "@ai how do I claim land?"
2. Bot waits for AI response
3. Bot logs response time and content
4. Bot 2 sends next question
5. Repeat through question list

Logging:
[2026-02-11 14:32:10] Bot1 -> @ai how do I claim land?
[2026-02-11 14:32:13] AI -> Use /claim while standing in area... (3.2s)
[2026-02-11 14:32:15] Bot2 -> @ai what's the economy?
[2026-02-11 14:32:18] AI -> We use coins as currency... (2.9s)

Test: 100 AI queries, measure response time

Configuration:
- 10 bots
- Each sends 10 questions
- Questions sent sequentially (wait for response)

Results:
Questions: 100
Successful: 98
Failed: 2 (timeout)
Avg response time: 4.2s
Median: 3.8s
95th percentile: 8.1s
Max: 12.4s

Interpretation:
✅ Good: 98% success rate
⚠️ Warning: 2 timeouts (investigate AI service)
✅ Good: Average under 5s
⚠️ Acceptable: 95th percentile under 10s (some slow queries)

Test: Concurrent query load

Configuration:
- 50 bots
- All send @ai query within 1 second

Monitor:
- How many queries answered?
- Response time distribution
- Server TPS impact
- AI service CPU/RAM

Results:
Queries sent: 50
Queries answered: 50
Time to answer all: 45 seconds
TPS during test: 18-20 (acceptable)
AI service CPU: 85% (near capacity)

Interpretation:
✅ All queries answered (queue working)
⚠️ 45 seconds for 50 queries = ~1s per query (queueing adds delay)
⚠️ High CPU on AI service (may need scaling)