storyteller/docs/reference/LLM_GUIDE.md

🤖 LLM Model Guide

Overview

Each character in the storyteller RPG can use a different AI model, giving them unique personality traits and response styles. This creates incredible variety and emergent gameplay!

Available Models

OpenAI Models (requires OPENAI_API_KEY)

GPT-4o (Latest)

• Best for: All-around excellence, latest capabilities
• Personality: Intelligent, balanced, reliable
• Cost:
• Speed: Fast

GPT-4 Turbo

• Best for: Complex reasoning, detailed responses
• Personality: Thoughtful, articulate, analytical
• Cost: $$$
• Speed: Medium

GPT-3.5 Turbo

• Best for: Quick interactions, budget-friendly
• Personality: Energetic, conversational, casual
• Cost: $
• Speed: Very Fast

OpenRouter Models (requires OPENROUTER_API_KEY)

Claude 3.5 Sonnet (Anthropic)

• Best for: Creative roleplay, nuanced responses
• Personality: Thoughtful, creative, emotionally aware
• Cost: $$$
• Speed: Fast
• Great for: Characters with depth and complexity

Claude 3 Opus (Anthropic)

• Best for: Most sophisticated responses
• Personality: Highly intelligent, philosophical
• Cost:
• Speed: Medium
• Great for: Wise characters, strategists

Claude 3 Haiku (Anthropic)

• Best for: Quick, concise responses
• Personality: Efficient, direct, clever
• Cost: $
• Speed: Very Fast
• Great for: Action-oriented characters

Gemini Pro 1.5 (Google)

• Best for: Factual knowledge, analytical thinking
• Personality: Logical, informative, precise
• Cost: $$
• Speed: Fast
• Great for: Scholarly characters, investigators

Llama 3.1 70B (Meta)

• Best for: Free-spirited, creative responses
• Personality: Bold, spontaneous, unpredictable
• Cost: $$
• Speed: Medium
• Great for: Wild characters, rogues

Llama 3.1 8B (Meta)

• Best for: Fast, lightweight interactions
• Personality: Quick-witted, energetic
• Cost: $
• Speed: Very Fast
• Great for: Nimble characters, scouts

Mistral Large (Mistral AI)

• Best for: European flair, multilingual
• Personality: Cultured, articulate, sophisticated
• Cost: $$
• Speed: Fast
• Great for: Noble characters, diplomats

Command R+ (Cohere)

• Best for: Following instructions precisely
• Personality: Obedient, structured, methodical
• Cost: $$
• Speed: Fast
• Great for: Soldiers, servants, loyal companions

Mixing Models for Rich Gameplay

Example Party Compositions

The Diverse Adventurers

• Wise Wizard: Claude 3 Opus (philosophical, strategic)
• Brave Warrior: GPT-4 Turbo (tactical, heroic)
• Sneaky Rogue: Llama 3.1 70B (unpredictable, bold)
• Scholar: Gemini Pro (analytical, knowledgeable)

The Quick Squad

• Scout: Llama 3.1 8B (fast, energetic)
• Fighter: Claude 3 Haiku (direct, efficient)
• Mage: GPT-3.5 Turbo (quick casting)

The Elite Team

• Leader: GPT-4o (balanced, excellent)
• Advisor: Claude 3.5 Sonnet (creative strategy)
• Specialist: Gemini Pro (expert knowledge)

Cost Considerations

Models are charged per token (roughly per word). Approximate costs:

• $: ~$0.50-1 per 1000 messages
• $$: ~$1-3 per 1000 messages
• $$$: ~$3-10 per 1000 messages
• $$$$: ~$10-30 per 1000 messages

Tips to save money:

• Use cheaper models (GPT-3.5, Claude Haiku, Llama 8B) for most characters
• Reserve expensive models (GPT-4o, Claude Opus) for key NPCs or special moments
• Mix and match based on character importance

Setting Up API Keys

OpenAI

1. Create account at https://platform.openai.com
2. Add payment method
3. Generate API key at https://platform.openai.com/api-keys
4. Add to .env: OPENAI_API_KEY=sk-...

OpenRouter

1. Create account at https://openrouter.ai
2. Add credits (starts at $5)
3. Generate API key at https://openrouter.ai/keys
4. Add to .env: OPENROUTER_API_KEY=sk-...

Why OpenRouter?

• Single API key for 100+ models
• Pay-as-you-go pricing
• No monthly subscriptions
• Access to Claude, Llama, Gemini, Mistral, and more

Testing Models

Try creating characters with different models and see how they respond differently to the same situation!

Example test:

1. Create 3 characters with different models
2. Have all three search a mysterious room
3. Compare their unique approaches:
   • GPT-4: Methodical, detailed search
   • Claude: Creative interpretations
   • Llama: Bold, risky actions
   • Gemini: Logical deductions

Advanced: Custom Model Configuration

You can add more models by editing main.py:

# In the get_available_models() function
models["openrouter"] = [
    {"id": "your/model-id", "name": "Display Name", "provider": "Provider"},
    # Add more models here
]

Find model IDs at https://openrouter.ai/models