AgentCore Memory 

AgentCore Memory 

Memory is a critical component of intelligence. While Large Language Models (LLMs) have impressive capabilities, they lack persistent memory across conversations. Amazon Bedrock AgentCore Memory addresses this limitation by providing a managed service that enables AI agents to maintain context over time, remember important facts, and deliver consistent, personalized experiences.

AgentCore Memory operates on two levels:

• Short-Term Memory: Immediate conversation context and session-based information that provides continuity within a single interaction or closely related sessions.

• Long-Term Memory: Persistent information extracted and stored across multiple conversations, including facts, preferences, and summaries that enable personalized experiences over time.

Each Memory ID (mem-28947293847293) is more like a "System-Level Container" that holds the entire memory structure for the application. The namespaces within it organize different users' data.

Memory ID: mem-28947293847293

    ├── support/customer/sarah_123/preferences/

    │   ├── email_preference

    │   └── language_preference

    │

    ├── support/customer/fred_123/preferences/

    │   ├── email_preference

    │   └── language_preference

    │

    ├── support/customer/sarah_123/semantic/

    │   └── past_conversations

    │

    └── support/customer/fred_123/semantic/

        └── past_conversations

Think of it like:

• Memory ID = Large Office Building - mem-28947293847293

• Namespaces = Different Floors/Sections

• User IDs (sarah_123, fred_123) = Individual Offices

• Preferences/Semantic = File Cabinets in each office

Real-world Example:

One Memory System (mem-28947293847293)

1. • Serves entire customer support system

   • Contains all users' data

   • Organized by namespaces

   • Further organized by user IDs

We now understand the memory id for the system ( large office building ) , in a long conversation across different calls or sessions we need to organize this data . when we create a memory with all or any  these strategy then in a chat conversation the framework extracts and puts the data in right namespaces by reasoning with chat and deciding strategy

StrategyType 

is from Amazon's Agent Core framework. main strategy types available:

USER_PREFERENCE
Stores user preferences
Example: Language choice, contact methods, time zones

SEMANTIC
Stores conversational facts and knowledge
      Example: Past issues, solutions, technical details

EPISODIC
Stores sequential events/interactions
Example: Conversation history in chronological order

DECLARATIVE
Stores explicit facts and statements
Example: Product specifications, policy information

PROCEDURAL
Stores step-by-step procedures
Example: Troubleshooting steps, workflows

The framework automatically processes conversations and distributes information to appropriate namespaces based on strategies. Here's how:

Simple Example Flow:

Customer: "I prefer to be contacted by email only and my working hours are 9-5 EST"
↓
Framework Analysis:
↓
Automatically Identifies:
- This is a preference → USER_PREFERENCE strategy
          - Stores in: support/customer/{id}/preferences/

Another Example:

Customer: "Last month I had issues with EC2 instances crashing"
↓
Framework Analysis:
↓
Automatically Identifies:
- This is historical fact → SEMANTIC strategy
          - Stores in: support/customer/{id}/semantic/

Real-world Scenario:

Conversation:
Customer: "Hi, I'm Sarah. I prefer emails, not calls.  Last week my database crashed at 3 AM."
↓
Automatic Distribution:

1. "prefer emails" → USER_PREFERENCE namespace

2. "database crashed" → SEMANTIC namespace

Think of it like a Smart Assistant:

• Listening to conversation

• Understanding context

• Filing information automatically

• Using right filing cabinet (strategy)

• Without manual intervention

The beauty is:

• No manual categorization needed

• Framework handles the reasoning

• Automatically sorts information

• Uses AI to understand context

• Places data in appropriate namespaces
  
  

The data in Agent Core memory is stored in vector stores and persistent storage managed by AWS. Here's the breakdown:

Storage Architecture:

Memory System

├── Vector Store (for semantic search) - Uses Amazon Bedrock's vector database

│   ├── Embeddings of conversations

│   └── Semantic information

│

├── Persistent Storage  Uses AWS managed services, survives system restarts

│   ├── User preferences

│   ├── Raw conversation data

│   └── Structured information

│

└── Temporary Cache (RAM)

    └── Active session data