Amazon Bedrock AgentCore Memory

Overview

AgentCore Memory enables agents to maintain persistent knowledge across sessions, learning from user interactions to provide increasingly personalized experiences. It combines short-term session context with long-term episodic memory extracted through background reflection processes.

Purpose: Give agents persistent memory and learning capabilities

Pattern: Capabilities-based (2 memory types)

Key Principles (validated by AWS December 2025):

1. Episodic Memory - Long-term facts extracted from conversations
2. Short-term Memory - Raw turn-by-turn session context
3. Automatic Extraction - Background reflection creates episodes
4. Semantic Retrieval - Context-aware memory lookup
5. User-Scoped - Memory isolated per user/actor
6. Privacy Controls - Granular memory management