Memory Management

Agent Kernel provides pluggable memory management capabilities.

Short-term Memory

Managed via Session objects for conversational context.

In-Memory Storage

export AK_SESSION__TYPE=in_memory

Use cases:

• Development
• Testing
• Single-process applications
• Non-critical data

Limitations:

• Lost on restart
• Single process only
• No persistence

Redis Storage

export AK_SESSION__TYPE=redis
export AK_SESSION__REDIS__URL=redis://localhost:6379
export AK_SESSION__REDIS__PASSWORD=your-password
export AK_SESSION__REDIS__TTL=3600  # 1 hour

Use cases:

• Production deployments
• Multi-process applications
• Distributed systems
• Session persistence required

Benefits:

• Persistent across restarts
• Shared across instances
• Configurable TTL
• High performance

DynamoDB Storage

export AK_SESSION__TYPE=dynamodb
export AK_SESSION__DYNAMODB__TABLE_NAME=agent-kernel-sessions
export AK_SESSION__DYNAMODB__TTL=3600  # 1 hour (0 to disable)

Use cases:

• AWS serverless deployments (Lambda)
• Auto-scaling requirements
• AWS-native infrastructure
• Serverless architectures

Benefits:

• Fully managed, serverless
• Auto-scaling capacity
• AWS-native integration
• No server maintenance
• Pay-per-use pricing

Requirements:

• DynamoDB table with partition key session_id (String)
• DynamoDB table with sort key key (String)
• Appropriate AWS IAM permissions
• Optional: TTL attribute expiry_time enabled on the table
• DynamoDB table and necessary permissions will be created automatically if you use the Agent Kernel provided terraform modules

Memory Architecture

Best Practices

Short-term Memory

• Use Redis in production
• Set appropriate TTL
• Monitor memory usage
• Clean up old sessions

# Configure TTL for Redis
export AK_SESSION__REDIS__TTL=7200  # 2 hours

# Configure TTL for DynamoDB
export AK_SESSION__DYNAMODB__TTL=7200  # 2 hours

Long-term Memory (Available soon!)

• Index frequently accessed data
• Implement data retention policies
• Back up important data
• Monitor storage costs

Summary

• Short-term memory for conversation context
• Long-term memory for persistent knowledge
• Redis recommended for containerized production deployments
• DynamoDB recommended for non-performance critical AWS serverless deployments
• Framework-specific long-term storage options
• Configurable TTL and retention
• Custom backends supported