mc-bot-town-2/nodejs/OLLAMA_SETUP.md

Ollama Integration Guide

This project now supports Ollama as an AI backend alongside Google Gemini, with per-bot configuration allowing you to mix providers and personalities across multiple bots.

Configuration Hierarchy

AI settings are merged in this order:

1. Global defaults in conf/base.js → ai object
2. Bot-specific overrides in conf/secrets.js → mc.bots.{botName}.plugins.Ai

Configuration

Global Defaults (Optional)

Edit conf/base.js to set defaults for all bots:

"ai":{
    // Default provider (can be overridden per-bot)
    "provider": "gemini",  // or "ollama"

    // Gemini API key (used by Gemini provider)
    "key": "<configure in conf/secrets.js>",

    // Ollama settings (used by Ollama provider)
    "baseUrl": "http://localhost:11434",
    "model": "llama3.2",
    "timeout": 30000,

    // Generation settings (applies to both providers)
    "temperature": 1,
    "topP": 0.95,
    "topK": 64,
    "maxOutputTokens": 8192,
    "interval": 20,
    // ... prompts
}

Per-Bot Configuration

Edit conf/secrets.js to configure each bot individually:

Example 1: Bot using default global settings

"art": {
    "username": "art@vm42.us",
    "commands": ['fun', 'invite', 'default'],
    "auth": "microsoft",
    "plugins": {
        "Ai":{
            "promptName": "helpful",
            // Uses global provider settings from base.js
        }
    },
},

Example 2: Bot using specific Ollama instance

"ayay": {
    "username": "limtisengyes@gmail.com",
    "commands": ['fun', 'invite', 'default'],
    "auth": "microsoft",
    "plugins": {
        "Ai":{
            "promptName": "asshole",
            "provider": "ollama",
            "baseUrl": "http://192.168.1.50:11434",  // Remote Ollama
            "model": "llama3.2:7b",
            "interval": 25,
        }
    }
},

Example 3: Bot using Gemini with custom settings

"nova": {
    "username": "your@email.com",
    "auth": "microsoft",
    "commands": ['default', 'fun'],
    "plugins": {
        "Ai":{
            "promptName": "helpful",
            "provider": "gemini",
            "model": "gemini-2.0-flash-exp",
            "temperature": 0.7,
        }
    }
},

Multiple Bots with Different Providers

You can run multiple bots with different providers simultaneously:

// conf/secrets.js
"bots": {
    "bot1": {
        "plugins": {
            "Ai": {
                "promptName": "helpful",
                "provider": "gemini",  // Uses Google Gemini
                "model": "gemini-2.0-flash-exp",
            }
        }
    },
    "bot2": {
        "plugins": {
            "Ai": {
                "promptName": "asshole",
                "provider": "ollama",    // Uses local Ollama
                "baseUrl": "http://localhost:11434",
                "model": "llama3.2",
            }
        }
    },
    "bot3": {
        "plugins": {
            "Ai": {
                "promptName": "Ashley",
                "provider": "ollama",    // Uses remote Ollama
                "baseUrl": "http://192.168.1.50:11434",
                "model": "mistral",
            }
        }
    }
}

Mixing Personalities and Models

Each bot can have:

• Different provider (Gemini or different Ollama instances)
• Different model (llama3.2, mistral, qwen2.5, etc.)
• Different personality (helpful, asshole, Ashley, custom)
• Different settings (temperature, interval, etc.)

"helpfulBot": {
    "plugins": {
        "Ai": {
            "promptName": "helpful",
            "provider": "ollama",
            "baseUrl": "http://server1:11434",
            "model": "llama3.2:3b",
            "temperature": 0.5,
            "interval": 15,
        }
    }
},
"toxicBot": {
    "plugins": {
        "Ai": {
            "promptName": "asshole",
            "provider": "ollama",
            "baseUrl": "http://server2:11434",
            "model": "llama3.2:70b",
            "temperature": 1.2,
            "interval": 30,
        }
    }
},

Ollama Setup

Install Ollama

# Linux/macOS
curl -fsSL https://ollama.com/install.sh | sh

# Or download from https://ollama.com/download

Pull Models

# Recommended for chat bots:
ollama pull llama3.2
ollama pull mistral
ollama pull qwen2.5

# Specific sizes for performance tuning:
ollama pull llama3.2:3b   # Fast, lightweight
ollama pull llama3.2:7b   # Good balance
ollama pull llama3.2:70b  # Smarter, slower

Start Ollama Server

# Local (only)
ollama serve

# Allow remote connections (for multiple servers)
OLLAMA_HOST=0.0.0.0:11434 ollama serve

Configure Remote Ollama

To use Ollama on another machine:

1. On the Ollama server:

   OLLAMA_HOST=0.0.0.0:11434 ollama serve
   

2. In bot config:

   "Ai": {
       "provider": "ollama",
       "baseUrl": "http://ollama-server-ip:11434",
       "model": "llama3.2",
   }
   

Ollama Model Recommendations

Model	Size	Speed	Quality	Best For
llama3.2:3b	3B	Very Fast	Good	Bots needing fast responses
llama3.2:7b	7B	Fast	Very Good	General purpose
llama3.2:70b	70B	Moderate	Excellent	Smart bots, complex prompts
mistral	7B	Fast	Good	Balanced solution
qwen2.5:7b	7B	Fast	Very Good	Good instruction following
gemma2:9b	9B	Fast	Good	Lightweight alternative

Troubleshooting

Connection Refused

# Check if Ollama is running
curl http://localhost:11434/api/tags

# Check specific server
curl http://192.168.1.50:11434/api/tags

Model Not Found

# Check available models
ollama list

# Pull missing model
ollama pull llama3.2

JSON Parsing Errors

Most models support JSON mode. If issues occur:

1. Switch to llama3.1, qwen2.5, or mistral
2. Lower temperature: (e.g., 0.7)
3. Increase maxOutputTokens: for longer responses

Slow Responses

• Use smaller models (llama3.2:3b vs 70b)
• Increase interval: in config
• Reduce maxOutputTokens:
• Check network latency for remote Ollama instances

Multiple Bots Overloading Ollama

If running many bots on one Ollama server:

1. Use lighter models for less important bots
2. Increase interval: to space requests
3. Distribute bots across multiple Ollama instances

Available Personality Prompts

Personality	Description	Best Model
helpful	Shy, helpful Jimmy	llama3.2, mistral
asshole	Sarcastic, unfiltered	llama3.2:70b, gemini
Ashley	Adult content	llama3.2, gemini
custom	Template for custom prompts	Any

Comparing Providers

Feature	Gemini	Ollama
Cost	API cost	Free (local)
Latency	200-500ms	50-500ms (local)
Privacy	Cloud	100% local
Multiple Servers	No	Yes
Model Choice	Limited	Any
Hardware	None Required	GPU Recommended
Offline	No	Yes

Command Reference

/msg botname ai <personality>     # Change personality
/msg botname ai <personality> custom message  # Use custom prompt
/msg wmantly load botname Ai <personality>  # Reload AI with new config