Part 10 — Architecture Patterns for Agent Memory

This section compares common architecture patterns for agent memory — from context-only systems to tiered stores, graphs, and enterprise context layers — and when to choose each.

What you will learn

  • Canonical patterns from no-persistence through managed memory services
  • Single-store vs multi-store and layered working/session/long-term designs
  • How topics, groups, and templates organize extraction configuration
  • Trade-offs between flexibility and simplicity when picking a pattern

Chapters in this section

There are 18 chapters in this part. Open any chapter to read it on its own, or work through them in order.

  1. 1 Pattern One: No Persistent Memory (Context-Only Systems)
  2. 2 Pattern Two: Flat Vector Store Memory
  3. 3 Pattern Three: Tiered Memory Architectures
  4. 4 Pattern Four: Graph-Based Memory Architectures
  5. 5 Pattern Five: The Enterprise Context Layer
  6. 6 Combining Experiential Memory with Organizational Knowledge
  7. 7 Single-Store vs Multi-Store Memory Architectures
  8. 8 Choosing an Architecture Pattern by Use Case
  9. 9 Memory Architecture for Single-Agent Systems
  10. 10 Memory Architecture for Multi-Agent Systems
  11. 11 Hybrid Architectures: Vector Store Plus Knowledge Graph
  12. 12 Layered Memory: Working, Session, and Long-Term Tiers
  13. 13 Designing for Adaptability: Avoiding Prescriptive Memory Models
  14. 14 Topics as a Configuration Layer for Memory Extraction
  15. 15 Groups: Bundling Configuration by Use Case
  16. 16 Templates vs Fully Custom Memory Pipelines
  17. 17 When to Build vs When to Use a Managed Memory Service
  18. 18 Trade-offs Between Flexibility and Simplicity in Memory Design