AI Services

Create and manage AI services for LLMs and embeddings using your configured providers

What Are AI Services?

AI Services are specific AI model instances that you configure for use in your workflows. While AI Providers establish connections to external AI platforms, AI Services define the actual models, settings, and configurations that power your AI features.

Prerequisites: You must have at least one AI Provider configured before creating AI Services. See the AI Providers Overview for setup instructions.

Types of AI Services

Elementum supports two primary types of AI Services:

LLM Services

Language Models for text generation, conversation, and analysis

Used for: Agents, automation actions, data analysis, and content generation

Examples: GPT-4, Claude 3.5, Gemini Pro

Embedding Services

Embedding Models for semantic search and similarity analysis

Used for: AI Search, content similarity, and semantic understanding

Examples: Snowflake Arctic L V2.0, Snowflake Arctic L V1.5

Creating AI Services

Accessing AI Services

1. Navigate to Services

   In Organization Settings, go to the Services tab

2. Create New Service

   Click ”+ Service” to open the service creation dialog

3. Select Provider

   Choose from your configured AI Providers:

   • OpenAI - For general-purpose AI capabilities
   • Snowflake - For data-native AI on your warehouse
   • Gemini - For advanced multimodal AI features

Creating LLM Services

LLM Services power conversational AI, text generation, and intelligent automation:

Service Configuration

Service Name: Give your service a descriptive name (e.g., “Customer Support Bot”)

Provider: Select your configured AI Provider

Model: Choose from available models:

• OpenAI o4-mini - Fast, efficient reasoning for daily tasks
• OpenAI o3 - Complex reasoning and research tasks
• Claude Sonnet 4 - Advanced reasoning and premium applications
• Claude 3.7 Sonnet - Cost-effective reasoning for most tasks
• Claude Opus 4 - Most complex reasoning (expensive but capable)
• Gemini 2.5 - Balanced performance for general-purpose tasks
• Gemini 2.5 Pro - Complex use cases and large responses

Cost Per Million Tokens: Optional cost tracking (varies by provider)

Creating Embedding Services

Embedding Services enable AI Search and semantic understanding:

Service Configuration

Service Name: Descriptive name (e.g., “Document Search Embeddings”)

Provider: Select your configured AI Provider

Model: Choose from available embedding models:

• Snowflake Arctic L V2.0 - Latest high-quality embeddings
• Snowflake Arctic L V1.5 - Reliable embeddings for production use

Dimensions: Embedding vector size (varies by model)

Service Management

Testing Services

Before using AI Services in production, test them thoroughly:

1. Access Testing Interface

   In the Services list, click on your service name to open the testing interface

2. Test LLM Services

   Input: Enter sample prompts or questions

   Response: Review the AI-generated responses

   Parameters: Adjust settings and test again

   Performance: Monitor response times and quality

3. Test Embedding Services

   Input: Enter sample text for embedding

   Vector Output: Review generated embedding vectors

   Similarity: Test similarity calculations between texts

   Performance: Monitor embedding generation speed

Service Monitoring

Usage Metrics

Token Consumption: Track token usage across services

Request Volume: Monitor API call frequency

Response Times: Track performance metrics

Error Rates: Monitor service reliability

Cost Management

Cost Tracking: Monitor spending per service

Budget Alerts: Set up spending notifications

Optimization: Identify cost-saving opportunities

Usage Reports: Generate regular usage reports

Service Usage Across Features

LLM Services Usage

Conversational Agents

Purpose: Power chatbots and customer service agents

Configuration:

• Use models optimized for conversation (o4-mini, Claude 3.7 Sonnet, Gemini 2.5)
• Set appropriate temperature for natural responses
• Configure stop sequences for conversation control

Best Practices:

• Use o4-mini for most customer support interactions
• Use Claude 3.7 Sonnet for cost-effective daily tasks
• Use Gemini 2.5 for balanced performance
• Set reasonable token limits for responses
• Use system prompts to define agent behavior

Example Use Cases:

• Customer support chatbots
• Internal help desk agents
• Sales qualification bots
• Technical support assistants

Automation Actions

Purpose: Enable AI-powered automation workflows

Configuration:

• Use task-appropriate models (o4-mini for simple tasks, o3 for complex reasoning)
• Set lower temperature for consistent results
• Configure appropriate token limits

Best Practices:

• Use o4-mini for most automation tasks
• Use o3 for complex reasoning and analysis
• Use Claude Sonnet 4 for detailed analysis
• Use Gemini 2.5 Pro for complicated use cases
• Use deterministic settings for predictable results
• Monitor automation performance regularly

Example Use Cases:

• Document classification
• Email response generation
• Data analysis and summarization
• Content transformation

Content Generation

Purpose: Generate written content and documentation

Configuration:

• Use models with strong writing capabilities (Claude Sonnet 4, Gemini 2.5 Pro)
• Adjust temperature based on creativity needs
• Set appropriate token limits for content length

Best Practices:

• Use Claude Sonnet 4 for premium content creation
• Use Gemini 2.5 Pro for large, detailed responses
• Use o3 for complex content requiring deep analysis
• Use detailed prompts for better results
• Implement content review processes
• Monitor quality and consistency

Example Use Cases:

• Report generation
• Email drafting
• Documentation creation
• Content summarization

Embedding Services Usage

AI Search

Purpose: Enable semantic search across your data

Configuration:

• Use high-quality embedding models (Snowflake Arctic L V2.0)
• Configure appropriate chunk sizes for your data
• Set up normalization for better similarity calculations

Best Practices:

• Use Snowflake Arctic L V2.0 for new implementations
• Use Snowflake Arctic L V1.5 for stable production workloads
• Optimize chunk sizes for your use case
• Implement proper indexing strategies

Requirements:

• Works with Standard Snowflake tables only
• Requires embedding service configuration
• Needs appropriate search index setup

Example Use Cases:

• Document search and retrieval
• Customer support knowledge bases
• Product catalog search
• Policy and procedure lookup

Content Similarity

Purpose: Find similar content and detect duplicates

Configuration:

• Use consistent embedding models across comparisons
• Configure appropriate similarity thresholds
• Set up batch processing for large datasets

Best Practices:

• Use the same embedding model for all content
• Implement proper similarity scoring
• Monitor performance with large datasets

Example Use Cases:

• Duplicate detection
• Content recommendation
• Similar document finding
• Categorization assistance

Best Practices

Model Selection

LLM Selection

Most Complex Tasks: Use o3 or Claude Opus 4 for advanced reasoning and research

Premium Applications: Use Claude Sonnet 4 or Gemini 2.5 Pro for detailed analysis

Daily Tasks: Use o4-mini, Claude 3.7 Sonnet, or Gemini 2.5 for most applications

Cost-Sensitive: Use Claude 3.7 Sonnet or o4-mini for cost-effective operations

Embedding Selection

High Quality: Use Snowflake Arctic L V2.0 for best search results

Production Ready: Use Snowflake Arctic L V1.5 for stable, reliable performance

Data-Native: All embeddings run directly on your Snowflake data warehouse

Consistency: Use the same embedding model throughout your search system

Performance Optimization

LLM Optimization

Temperature Settings:

• Use 0.0-0.3 for deterministic tasks
• Use 0.4-0.7 for balanced creativity
• Use 0.8-1.0 for creative tasks

Token Management:

• Set appropriate max tokens for responses
• Monitor token usage for cost control
• Use truncation strategies for long inputs

Prompt Engineering:

• Use clear, specific prompts
• Provide examples for better results
• Implement system prompts for consistency

Cost Management

Monitor Usage

Track Consumption: Monitor token usage across all services

Set Budgets: Establish spending limits for each service

Usage Patterns: Analyze usage patterns to optimize costs

Regular Review: Conduct monthly cost reviews

Optimize Costs

Right-Size Models: Use appropriate models for tasks

Batch Processing: Process multiple requests together

Caching: Cache frequent responses and embeddings

Efficient Prompts: Use concise, effective prompts

Troubleshooting

Service Creation Failures

Symptoms: Cannot create new AI services

Common Causes:

• AI Provider not configured
• Invalid model selection
• Insufficient permissions

Solutions:

1. Verify AI Provider is properly configured
2. Check model availability for your provider
3. Ensure proper permissions are granted
4. Try creating with different model options

Poor Performance

Symptoms: Slow response times or quality issues

Common Causes:

• Inappropriate model selection
• Suboptimal configuration
• Network or provider issues

Solutions:

1. Review model selection for your use case
2. Optimize service configuration settings
3. Check provider status and network connectivity
4. Consider switching to different models

High Costs

Symptoms: Unexpected high token usage or costs

Common Causes:

• Inefficient prompts or queries
• Inappropriate model selection
• Excessive API calls

Solutions:

1. Review and optimize prompts
2. Use more cost-effective models where appropriate
3. Implement caching and batching
4. Monitor and analyze usage patterns

Advanced Configuration

Custom Model Settings

For specialized use cases:

1. Fine-tuning: Some providers support custom model fine-tuning
2. Custom Endpoints: Configure custom API endpoints for specialized deployments
3. Advanced Parameters: Use provider-specific advanced settings
4. Performance Tuning: Optimize for specific performance requirements

Multi-Provider Strategy

Redundancy

Failover: Configure multiple providers for reliability

Load Balancing: Distribute requests across providers

Cost Optimization: Route requests to most cost-effective provider

Feature Specialization: Use different providers for different capabilities

Hybrid Approach

LLM Diversity: Use different LLMs for different tasks

Embedding Consistency: Maintain consistent embedding models

Regional Deployment: Use region-specific providers

Compliance Requirements: Meet different regulatory needs

Next Steps

With your AI Services configured:

Enable AI Search

Use embedding services to power intelligent search

Build Agents

Create conversational AI assistants using your LLM services

Use AI Actions

Add AI capabilities to your automation workflows

Monitor Performance

Set up monitoring and optimization for your AI services

---

AI Services bridge the gap between AI Providers and your actual AI-powered features. Properly configured services ensure optimal performance, cost-effectiveness, and reliability for your AI workflows.