• A managed service that enables AI agents to store, retrieve, and maintain context across conversations, allowing them to remember user information, preferences, and interaction history for more coherent and personalized responses.

How Memory Works

• Memory Storage and Retrieval
  • The Memory service automatically captures relevant information from agent-user conversations and stores it for future use. When an agent needs context, it queries the memory to retrieve past interactions, user details, or learned facts. The system uses semantic search to find the most relevant memories based on the current conversation.

• Short-Term vs. Long-Term Memory
  • Short-term memory handles context within a single conversation session, maintaining coherence during extended dialogues. Long-term memory persists across multiple sessions, allowing agents to remember users and preferences over days, weeks, or months. You can configure retention policies for each memory type.
• Memory Organization
  • Memories are organized by conversation threads, user IDs, or custom categories you define. Each memory entry includes the content, metadata (timestamp, source), and relevance tags. The system automatically structures memories for efficient retrieval while maintaining logical relationships between related pieces of information.

Memory Types and Configuration

• Conversation Memory
  • Stores the history of interactions between agents and users, including questions, responses, and contextual details. This enables agents to reference previous parts of a conversation and maintain continuity. You can configure how many turns or how far back in a conversation to retain.
• User Profile Memory
  • Maintains persistent information about users, such as preferences, personal details, and behavior patterns. This allows agents to personalize interactions based on accumulated knowledge about individual users. Profile memories are automatically updated as agents learn more about users.
• Knowledge Memory
  • Stores factual information, learned concepts, and discovered data that agents can reference in future conversations. This includes research findings, verified facts, and domain-specific information that enhances agent expertise over time.
• Custom Memory Schemas
  • Define your own memory structures for specific use cases using JSON schemas. Create custom memory types for specialized data like project details, task lists, or domain-specific information. Configure validation rules and retention policies for each custom memory type.

Amazon Bedrock AgentCore Memory Implementation

• Memory Enablement
  • Enable memory for your agents through the Bedrock AgentCore console or API. Configure global memory settings including default retention periods, storage limits, and privacy controls. Connect memory to existing agents with minimal code changes using the provided SDKs.
• Memory Population
  • Agents automatically populate memory during conversations based on configured rules. You can also manually add memories through API calls for initial data loading or corrections. Set up webhooks to sync memory with external databases or CRM systems.
• Retrieval Configuration
  • Configure how agents query memory using relevance thresholds and search parameters. Set up filters to control which memories are retrieved based on context, time, or confidence scores. Optimize retrieval performance by adjusting the balance between recall accuracy and response time.
• Eviction Policies
  • Define rules for automatically removing old or irrelevant memories. Set up retention periods based on memory type, importance, or storage constraints. 

Amazon Bedrock AgentCore Memory Integration 

• Agent-Memory Integration
  • Agents automatically access memory during conversations through built-in memory functions. The system handles memory lookup and context injection without requiring explicit memory calls in your agent logic. Configure when and how often agents should check memory during conversations.
• Multi-Agent Memory Sharing
  • Share memories between different agents working on related tasks or serving the same users. Implement memory sharing policies to control what information is shared between agents. Use memory references to connect related information across different agent domains.
• External System Integration
  • Connect memory to external databases, CRM systems, or knowledge bases for enriched context. Implement bi-directional sync to keep memory consistent with external data sources. Use APIs to export memory data for analytics or backup purposes.

Amazon Bedrock AgentCore Memory Security 

• Data Encryption
  • All memories are encrypted at rest using AWS KMS and in transit using TLS. Implement customer-managed keys for additional control over encryption. Apply field-level encryption for particularly sensitive information within memories.
• Access Controls
  • Use IAM policies to control which agents can read, write, or modify specific memories. Implement user-level permissions for memories containing personal data. Set up audit logging for all memory access and modification events.
• Privacy Compliance
  • Configure automatic data expiration for privacy-sensitive information. Implement anonymization for memories containing personal identifiable information. Set up compliance features for GDPR, CCPA, and other privacy regulations.
• Data Isolation
  • Ensure memories are properly isolated between different users, organizations, or applications. Implement tenant isolation for multi-tenant deployments. Use separate memory stores for different security classification levels.

Amazon Bedrock AgentCore Memory Best Practices

• Memory Design
  • Design memory schemas that balance detail with retrieval efficiency. Create clear naming conventions for memory types and categories. Implement versioning for memory schemas to handle updates without data loss.
• Performance Optimization
  • Monitor memory storage growth and implement archiving for older conversations. Optimize retrieval patterns based on your specific use case requirements. Implement caching for frequently accessed memories to reduce latency.

• Quality Management
  • Regularly review memory contents for accuracy and relevance. Implement feedback mechanisms to correct inaccurate memories. Set up alerts for unusual memory patterns or potential data quality issues.
• Cost Management
  • Configure appropriate retention periods to balance utility with storage costs. Implement tiered storage for different memory types based on access frequency. Monitor memory usage metrics to optimize cost-efficiency.

Amazon Bedrock AgentCore Memory Pricing

• Usage-Based Model
  • Memory is priced as part of Amazon Bedrock AgentCore usage with pay-per-use pricing. You are charged based on the storage volume of memories and the number of read/write operations. There are no upfront costs or minimum commitments for using the Memory service.
• Cost Components
  • Pricing is included as part of Amazon Bedrock AgentCore usage costs based on overall service consumption. Additional costs may apply for advanced features like custom encryption keys or extended retention periods. All usage is visible through AWS Cost Explorer with detailed breakdowns.

Amazon Bedrock AgentCore Memory Cheat Sheet References:

https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/memory.html
https://aws.amazon.com/blogs/machine-learning/amazon-bedrock-agentcore-memory-building-context-aware-agents/