Skip to content

04-AgentCore-memory

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History
 
 

04-AgentCore-memory

README.md

Amazon Bedrock AgentCore Memory

Overview

Memory is a critical component of Agent intelligence. Large Language Models (LLMs) lack persistent memory across conversations. Amazon Bedrock AgentCore Memory addresses this by providing a managed service that enables AI agents to maintain relevant context across sessions, deliver personalized experiences and help the agent to learn over time.

Key Capabilities

  • Core Infrastructure: Serverless setup with built-in encryption and observability
  • Event Storage: Raw event storage (conversation history/checkpointing) with branching support
  • Strategy Management: Configurable extraction strategies (SEMANTIC, SUMMARY, USER_PREFERENCES, EPISODIC, SELF_MANAGED)
  • Memory Records Extraction: Automatic extraction of facts, preferences, and summaries based on configured strategies
  • Semantic Search: Vector-based retrieval of relevant memories using natural language queries

How AgentCore Memory Works

high_level_workflow

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.

Memory Architecture

  1. Conversation Storage: Complete conversations are saved in raw form for immediate access
  2. Strategy Processing: Configured strategies automatically analyze conversations in the background
  3. Information Extraction: Important data is extracted based on strategy types (typically takes ~1 minute)
  4. Organized Storage: Extracted information is stored in structured namespaces for efficient retrieval
  5. Semantic Retrieval: Natural language queries can retrieve relevant memories using vector similarity

Memory Strategy Types

AgentCore Memory supports five strategy types:

  • Semantic Memory: Stores factual information using vector embeddings for similarity search
  • Summary Memory: Creates and maintains conversation summaries for context preservation
  • User Preference Memory: Tracks user-specific preferences and settings
  • Episodic Memory: Captures meaningful interaction sequences with automatic episode detection, consolidation, and reflection generation
  • Self-managed Memory: Allows customization of extraction and consolidation logic

Folder Structure

04-AgentCore-memory/
├── 01-short-term-memory/          # Session-based memory and context management
│   ├── 01-single-agent/
│   │   ├── with-strands-agent/    # Strands SDK examples + checkpointing
│   │   ├── with-langgraph-agent/  # LangGraph examples + checkpointing + human-in-the-loop
│   │   └── with-llamaindex-agent/ # LlamaIndex examples across multiple domains
│   └── 02-multi-agent/
│       └── with-strands-agent/    # Multi-agent travel planning
├── 02-long-term-memory/           # Persistent memory across conversations
│   ├── 01-single-agent/
│   │   ├── using-strands-agent-hooks/         # Strands lifecycle hooks integration
│   │   ├── using-strands-agent-memory-tool/   # Strands memory tool integration
│   │   ├── using-langgraph-agent-hooks/       # LangGraph hooks integration
│   │   └── using-llamaindex-agent-memory-tool/ # LlamaIndex memory tool integration
│   └── 02-multi-agent/
│       └── with-strands-agent/    # Multi-agent travel booking + healthcare
├── 03-advanced-patterns/          # Advanced integrations and tooling
│   ├── 01-guardrails-integration/ # Memory with Amazon Bedrock Guardrails
│   ├── 02-memory-runtime-integration/          # Memory + AgentCore Runtime
│   ├── 03-memory-identity-runtime-integration/ # Memory + Identity + Runtime
│   ├── 04-memory-browser/         # Web UI for browsing memory stores
│   └── 05-memory-streaming/       # Streaming memory record extraction
├── 04-memory-branching/           # Conversation branching and parallel execution
└── 05-memory-security-patterns/   # IAM policies and Cognito identity integration
    ├── 01-memory-iam-policies/
    └── 02-memory-iam-cognito-identities/

Sample Notebooks

Short-Term Memory

Framework Agent Type Use Case Notebook
Strands Single Personal Agent personal-agent.ipynb
Strands Single Personal Agent (Memory Manager) personal-agent-memory-manager.ipynb
LangGraph Single Personal Fitness Coach personal-fitness-coach.ipynb
LangGraph Single Math Agent with Checkpointing math-agent-with-checkpointing.ipynb
LangGraph Single Support Agent (Human-in-the-Loop) support-agent-human-in-the-loop.ipynb
LlamaIndex Single Academic Research Assistant academic-research-assistant.ipynb
LlamaIndex Single Investment Portfolio Advisor investment-portfolio-advisor.ipynb
LlamaIndex Single Legal Document Analyzer legal-document-analyzer.ipynb
LlamaIndex Single Medical Knowledge Assistant medical-knowledge-assistant.ipynb
Strands Multi Travel Planning Agent travel-planning-agent.ipynb
Strands Multi Travel Planning (Memory Manager) travel-planning-agent-memory-manager.ipynb

Long-Term Memory

Framework Agent Type Integration Use Case Notebook
Strands Single Hooks Customer Support (Built-in Strategy) customer-support-inbuilt-strategy.ipynb
Strands Single Hooks Customer Support (Override Strategy) customer-support-override-strategy.ipynb
Strands Single Hooks Math Assistant math-assistant.ipynb
Strands Single Hooks Meeting Notes (Episodic) meeting-notes-assistant.ipynb
Strands Single Hooks Culinary Assistant (Self-Managed) agentcore_self_managed_memory_demo.ipynb
Strands Single Hooks Culinary Assistant (Self-Managed + Citations) agentcore_self_managed_memory_demo.ipynb
Strands Single Memory Tool Culinary Assistant culinary-assistant.ipynb
Strands Single Memory Tool Debugging Assistant (Episodic) debugging_assistant_episodic_memory.ipynb
LangGraph Single Hooks Nutrition Assistant (User Preferences) nutrition-assistant-with-user-preference-saving.ipynb
LangGraph Single Hooks Nutrition Assistant (Episodic) nutrition-assistant-with-episodic-memory.ipynb
LlamaIndex Single Memory Tool Academic Research Assistant academic-research-assistant.ipynb
LlamaIndex Single Memory Tool Investment Portfolio Advisor investment-portfolio-advisor.ipynb
LlamaIndex Single Memory Tool Legal Document Analyzer legal-document-analyzer.ipynb
LlamaIndex Single Memory Tool Medical Knowledge Assistant medical-knowledge-assistant.ipynb
Strands Multi Hooks Travel Booking Assistant travel-booking-assistant.ipynb
Strands Multi Hooks Healthcare Data Assistant (Episodic) healthcare-data-assistant.ipynb

Advanced Patterns

Pattern Description Notebook
Guardrails Integration Combine memory with Amazon Bedrock Guardrails guardrails-memory.ipynb
Memory + Runtime Integrate memory with AgentCore Runtime runtime_memory_integration.ipynb
Memory + Identity + Runtime Integrate memory, identity resolution, and runtime runtime_memory_identity_integration.ipynb
Memory Browser Web UI for exploring and managing memory stores README
Memory Streaming Stream memory record extraction results memory_record_streaming.ipynb

Memory Branching

Use Case Notebook
Travel Planning with Memory Branching travel-planning-agent-with-memory-branching.ipynb
Multi-Agent Parallel Execution with Branching multi-agent-parallel-execution-with-memory-branching.ipynb

Memory Security Patterns

Pattern Notebook
IAM Policies for Memory Access Control runtime_memory_identity_integration.ipynb
IAM + Cognito Federated Identities runtime_memory_federated_identity_integration.ipynb

Resources

Prerequisites

  • Python 3.10 or higher
  • AWS account with Amazon Bedrock access
  • Jupyter Notebook environment
  • Required Python packages (see individual sample requirements.txt files)