Model Configuration

Model Configuration Guide

VaultAI supports fine-tuning AI responses through advanced model configuration parameters. These settings allow you to control the creativity, randomness, and length of generated content.

Configuration Parameters

Temperature (0.0 - 1.0)

Controls the randomness of AI responses.

• 0.0: Deterministic, always chooses most likely response
• 0.3-0.5: Balanced, good for factual content
• 0.7-0.9: Creative, good for writing and brainstorming
• 1.0: Maximum creativity and randomness

Default: 0.7

Top-K (1 - 40)

Limits the AI to consider only the top K most probable tokens at each step.

• 1-10: Very focused responses
• 20-30: Balanced variety (recommended)
• 40: Maximum vocabulary diversity

Default: 40

Top-P (0.0 - 1.0)

Nucleus sampling - considers tokens that make up the top P probability mass.

• 0.1-0.3: Very focused
• 0.7-0.9: Balanced (recommended)
• 1.0: Considers all tokens

Default: 0.9

Max Output Tokens (1 - 8192)

Maximum length of generated responses.

• 512: Short responses
• 1024: Medium responses
• 2048: Long responses (recommended)
• 4096-8192: Very long responses

Default: 2048

How to Configure

In Obsidian Plugin Settings

1. Open Settings → Community Plugins → VaultAI
2. Navigate to Model Configuration section
3. Adjust the sliders or enter values directly
4. Settings apply immediately to new conversations

Recommended Settings by Use Case

📝 Note-Taking & Summaries

• Temperature: 0.3
• Top-K: 20
• Top-P: 0.8
• Max Tokens: 1024

✍️ Creative Writing

• Temperature: 0.9
• Top-K: 40
• Top-P: 0.95
• Max Tokens: 2048

🤔 Analysis & Research

• Temperature: 0.5
• Top-K: 30
• Top-P: 0.9
• Max Tokens: 2048

💡 Brainstorming

• Temperature: 0.8
• Top-K: 35
• Top-P: 0.9
• Max Tokens: 1024

Tips & Best Practices

1. Start Conservative: Begin with lower temperature (0.5-0.7) and adjust based on results
2. Experiment: Different tasks benefit from different settings
3. Balance Parameters: High temperature with low top-K can produce inconsistent results
4. Token Limits: Higher token limits allow for more detailed responses but use more API quota
5. Save Presets: Create custom presets for different types of work

Troubleshooting

Responses Too Random/Nonsensical

• Lower temperature (try 0.3-0.5)
• Reduce top-K (try 20)
• Lower top-P (try 0.8)

Responses Too Repetitive/Boring

• Increase temperature (try 0.7-0.9)
• Increase top-K (try 30-40)
• Increase top-P (try 0.95)

Responses Cut Off

• Increase max output tokens
• Break complex requests into smaller parts

API Quota Usage Too High

• Reduce max output tokens
• Use more focused prompts
• Consider using lower temperature for factual queries

Advanced Configuration

For power users, VaultAI also supports:

• Model Selection: Choose between different Gemini models
• Context Window: Adjust how much conversation history to include
• Streaming: Enable real-time response streaming
• Custom System Prompts: Set behavior guidelines for the AI

See Advanced Settings for more details.

---

Need Help? Check out our Troubleshooting Guide or join the Community Discussions.