Overview
Snowflake Cortex brings AI capabilities directly to your data warehouse, allowing you to run LLMs and embedding models on your data without moving it outside your secure environment. This setup leverages your existing CloudLink connection to access Snowflake’s AI capabilities.Prerequisites: You must have a CloudLink connection configured with key-pair authentication to access Snowflake Cortex AI features.
Authentication Requirement: Snowflake Cortex AI features are only available when using key-pair authentication. Password authentication cannot access these capabilities.
Step 1: Verify CloudLink Prerequisites
Check Your CloudLink Connection
Before setting up Snowflake Cortex, ensure your CloudLink is properly configured:-
Navigate to CloudLink Settings
- Go to Organization Settings → Cloud Credentials
- Verify your Snowflake connection is active
-
Confirm Key-Pair Authentication
- Your CloudLink must use key-pair authentication
- If using password authentication, you’ll need to upgrade to key-pair
Verify Snowflake Account Access
Your Snowflake account must have access to Cortex AI features:Account Requirements
Snowflake Edition: Enterprise or higherCortex Features: Available in most regionsPermissions: USAGE privileges on Cortex functions
Region Availability
Supported Regions: Most AWS, Azure, and GCP regionsModel Access: Varies by region and account typeBilling: Cortex usage billed through Snowflake
Step 2: Enable Snowflake Cortex in Elementum
Automatic Discovery
When you have a CloudLink connection with key-pair authentication, Elementum automatically discovers available Snowflake Cortex capabilities:1
Navigate to Providers
In Organization Settings, go to the Providers tab
2
Create Snowflake Provider
Click ”+ Provider” and select SnowflakeYou’ll see your existing CloudLink connections listed
3
Configure Provider
Provider Name: Enter a descriptive name (e.g., “Snowflake Cortex AI”)CloudLink: Select your key-pair authenticated CloudLinkService Account Credentials: Auto-populated from your CloudLink
4
Save and Test
Click “Save” to create the providerElementum will test the connection and discover available models
Manual Configuration
If automatic discovery doesn’t work:- Connection Details
- Service Account
Provider Name: Descriptive name for your Snowflake providerLocation: Your Snowflake region and account detailsProject ID: Your Snowflake account identifierCloudLink: Select the appropriate CloudLink connection
Step 3: Available Snowflake Cortex Models
Once configured, you’ll have access to various AI models through Snowflake Cortex:Language Models (LLMs)
| Model | Primary Use Case | Speed | Intelligence | Best For |
|---|---|---|---|---|
| Claude Sonnet 4 | Advanced reasoning and analysis | ⭐⭐⭐ | ⭐⭐⭐⭐⭐ | Complex problem-solving, detailed analysis, premium applications |
| Claude 3.7 Sonnet | Cost-effective reasoning | ⭐⭐⭐⭐ | ⭐⭐⭐⭐ | Daily tasks, customer support, balanced performance |
| Claude Opus 4 | Most complex reasoning | ⭐⭐ | ⭐⭐⭐⭐⭐ | Extremely complex tasks, research, advanced analysis (expensive) |
| Mistral Large 2 | European AI compliance | ⭐⭐⭐ | ⭐⭐⭐⭐ | European regulations, multilingual tasks |
Model Recommendations: Use Claude 3.7 Sonnet for most daily tasks and cost-effective operations. Choose Claude Sonnet 4 for advanced reasoning and premium applications. Reserve Claude Opus 4 for the most complex tasks that require maximum intelligence (note: very expensive).
Embedding Models
| Model | Primary Use Case | Speed | Quality | Best For |
|---|---|---|---|---|
| Snowflake Arctic L V2.0 | Latest high-quality embeddings | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ | Modern search applications, premium AI Search |
| Snowflake Arctic L V1.5 | Reliable embeddings | ⭐⭐⭐⭐ | ⭐⭐⭐⭐ | Stable search applications, production use |
Embedding Recommendations: Use Snowflake Arctic L V2.0 for new implementations and highest quality search results. Arctic L V1.5 provides reliable performance for production workloads.
Model Availability: Available models depend on your Snowflake account tier, region, and current Cortex offerings. Model selection may vary over time.
Step 4: Create AI Services
With Snowflake Cortex configured, create AI Services to use the models:1
Navigate to Services
In Organization Settings, go to the Services tab
2
Create LLM Service
Click ”+ Service” and select from available Snowflake modelsService Name: Descriptive name (e.g., “Snowflake Claude 3.5”)Model: Select from discovered Cortex modelsCost Tracking: Set cost per million tokens (optional)
3
Create Embedding Service
For AI Search capabilities, create an embedding serviceService Name: Descriptive name (e.g., “Snowflake Embeddings”)Model: Select an embedding modelConfiguration: Additional settings for embedding behavior
4
Test Services
Use the built-in testing interface to verify services work correctly
Benefits of Snowflake Cortex
Data Security and Privacy
Data Residency
No Data Movement: AI runs directly on your data warehouseCompliance: Maintains data governance and complianceSecurity: Leverages Snowflake’s security modelPrivacy: Your data never leaves your environment
Performance
Direct Access: No data transfer latencyScalability: Leverages Snowflake’s compute powerOptimization: Models optimized for structured dataCost Efficiency: Reduces data movement costs
Integration Advantages
- Seamless Data Access: AI models can directly query your tables
- Real-Time Processing: Process data as it arrives
- Unified Platform: Single environment for data and AI
- Cost Optimization: Leverage existing Snowflake infrastructure
Configuration Best Practices
Performance Optimization
- Warehouse Sizing
- Data Preparation
Model Complexity: Larger models may require bigger warehousesConcurrent Usage: Scale warehouses based on simultaneous requestsCost Balance: Balance performance needs with compute costsAuto-Scaling: Enable auto-scaling for variable workloads
Cost Management
Monitor Usage
Cortex Billing: Monitor AI function usage in SnowflakeCompute Costs: Track warehouse usage for AI workloadsData Transfer: Minimize unnecessary data movementOptimization: Regularly review and optimize usage patterns
Efficiency Tips
Batch Processing: Process multiple requests togetherCaching: Cache frequent AI resultsModel Selection: Choose appropriate models for tasksQuery Optimization: Optimize SQL queries using AI functions
Troubleshooting
Cortex Functions Not Available
Cortex Functions Not Available
Symptoms: Cannot access Snowflake Cortex AI functionsCommon Causes:
- Using password authentication instead of key-pair
- Insufficient permissions on Cortex functions
- Account doesn’t have Cortex access
- Verify key-pair authentication is configured
- Check USAGE privileges on Cortex functions
- Contact Snowflake support for account access
- Verify account edition and region support
Model Discovery Issues
Model Discovery Issues
Symptoms: Expected models don’t appear in service creationCommon Causes:
- Regional model availability
- Account tier limitations
- CloudLink connection issues
- Verify CloudLink connection is active
- Check regional model availability
- Review account tier and permissions
- Refresh provider configuration
Performance Issues
Performance Issues
Symptoms: Slow AI response times or timeoutsCommon Causes:
- Undersized warehouse for AI workloads
- Inefficient query patterns
- Large data volumes
- Scale up warehouse size
- Optimize data queries
- Implement result caching
- Use appropriate table types
Advanced Configuration
Custom Model Access
For specialized or private models:- Model Registration: Register custom models in Snowflake
- Access Control: Configure proper permissions
- Performance Tuning: Optimize for your specific use case
- Monitoring: Set up custom monitoring and alerting
Multi-Region Setup
For global deployments:- Region Selection: Choose regions closest to your data
- Data Replication: Consider data residency requirements
- Failover: Implement failover strategies
- Compliance: Ensure regional compliance requirements
Next Steps
With Snowflake Cortex configured:Create AI Services
Set up specific LLM and embedding services using Cortex models
Enable AI Search
Use Snowflake embeddings for intelligent search on your data
Build Agents
Create agents that can directly access your Snowflake data
Optimize Performance
Configure warehouses for optimal AI performance
Snowflake Cortex provides secure, high-performance AI capabilities directly on your data warehouse. This approach ensures data privacy while delivering advanced AI features for your workflows.