This section treats forgetting as a design feature: importance, decay, pruning, summarization, and consolidation so memory stays useful instead of accumulating noise forever.
What you will learn
- Why remembering everything is expensive and often harmful
- Importance, recency, and decay models for retention
- Soft forgetting, archival, supersession, and contradiction handling
- Maintenance jobs that keep long-running memory stores healthy
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 Why Forgetting Is a Feature, Not a Bug
- 2 The Cost of Remembering Everything
- 3 Importance Scoring: Deciding What’s Worth Keeping
- 4 Recency and Access Frequency as Retention Signals
- 5 Exponential Decay Models for Memory Relevance
- 6 Soft Forgetting: Decay in Accessibility, Not Deletion
- 7 Hard Deletion vs Archival
- 8 Pruning Stale and Superseded Memories
- 9 Detecting Contradictions Between Old and New Memories
- 10 Superseding Memories Without Losing History
- 11 Summarization as a Memory Compression Strategy
- 12 Consolidation: Turning Episodic Memory Into Semantic Memory
- 13 Reconsolidation: Updating Memory Upon Retrieval
- 14 Handling Preference Drift Over Time
- 15 Memory Maintenance Jobs: Scheduled Cleanup and Merging
- 16 Staleness Detection in Long-Running Agent Deployments
- 17 Balancing Stability and Plasticity in Memory Systems
- 18 When to Let an Agent Forget on Purpose