AI agents

Designing agent memory: what to store and forget

• Memory for AI Agents: What to Store and What to Forget
• A Memory Write Policy for Agents: What to Record and Drop
• Compacting Agent Memory Without Losing the Facts That Matter
• Compaction Triggers: When a Fold Breaks the Answer
• Measuring the Cost of a Missing Fact: Recovery, Refusal, Re-Ask
• Merging Conflicting Chunks Before the Fold
• Replay Questions: A Test Set That Proves Memory Kept Facts
• Splitting a Memory Store by Trust: Read-Only, Draft, Quarantined
• Summarizer Prompts That Preserve Numbers, Names and Negations
• Testing a Memory Write Policy: Replay, Fixtures and Drift Checks
• A Fact Ledger for Agent Memory: Recording What Must Never Be Lost
• A Fold Regression Test: Fixtures That Catch Lost Facts
• A Provenance Field for Every Memory Write: Source and Confidence
• A Store Interface for Agent Memory Writes
• Choosing a Memory Backend: Postgres, Redis and Vector Stores
• Consistency Contracts for Agent Memory Stores
• Detecting a Lost Fact After a Fold: A Coverage Check
• Fold Rules in Versioned Config: Owner, Test and Rollback
• Folded Summaries vs Raw Facts: What to Retrieve First
• Scoping Agent Memory to a Tenant: Partitioning and Deletion
• Soft Delete, Hard Delete and Grace Windows for Agent Memory
• Storing Facts vs Summaries: What an Agent Retrieves First
• The Compaction Log: Fields That Prove What the Model Saw
• What a Memory Fold Costs in Tokens and Latency
• When an Agent Writes a Fact About a Person: Consent and Fixes