Retrieval and memory for AI agents
What agent memory should store
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
- All of What agent memory should store
Embedding and reranker models
Choosing embedding and reranking models
- Embedding Model Selection: Dimensions, Cost and Domain Fit
- When a Reranker Pays for Itself Against a Bigger Cap
- A Provider Interface for Embeddings: Batching and Retries
- A Reranker Is a Model Call: Budgeting Latency and Tokens
- Choosing an Embedding Model: Dimensions, Cost and Domain Fit
- Embedding a Tenant's Data: Where Vectors and Text Live
- Reranker Latency and Token Budget: A Per-Query Cost Model
- Reranker Versus a Bigger k: Which Spends the Budget Better
- All of Embedding and reranker models
Chunking and indexing
Chunking and indexing strategies
- Chunking Strategies for Retrieval: Size, Overlap and Boundaries
- Ordering Retrieved Chunks for Cache Hits
- Choosing a Pass Threshold for Recall After a Refresh
- Fact-Boundary Chunking: One Claim per Chunk
- Index Refresh: Rebuilding and Re-Embedding When Documents Change
- Re-Embedding After Index Refresh: Batching, Cost and Cutover
- Re-Embedding vs Re-Chunking: Which Refresh Do You Need?
- Span IDs as the Primary Key After a Re-Chunk
- All of Chunking and indexing
Hybrid search and context budgets
Hybrid search, reranking, and context budgets
- Retrieval-Augmented Generation Explained for Builders
- A Rewrite Budget per Query: How Many Facts Before Retrieval
- Citing Spans, Not Chunks: The Gate Gets Sharper
- Context Window Budget: How Many Chunks to Send and When to Stop
- Extractive Answers: Sending the Passage, Not the Document
- How Many Chunks to Send: A Context Window Budget for Answers
- Hybrid Search and Reranking: When and How to Combine Them
- Mapping Retrieval Budgets to Answer Quality Targets
- All of Hybrid search and context budgets
Debugging retrieval quality
Debugging and measuring retrieval quality
- Detecting a Citation the Agent Never Used
- Measuring Retrieval Quality: Hit Rate, MRR and Recall@k for RAG
- Measuring Retrieval Recall After a Document Refresh
- Provenance Fields That Survive a Regeneration
- Reading a Retrieval Change as a Curve Shift, Not a Score
- Scoring Retrieval With No Labels: Clicks and Citations
- What Recall@k Misses: A Complementary Retrieval Check
- A Citation Check That Skips the Model
- All of Debugging retrieval quality
Memory expiry and staleness
Memory expiry, invalidation, and staleness
- A Freshness Gate Before the Prompt: Fields and Refusals
- A Quarantine Table for Suspect Chunks: Schema and Retention
- Contract Tests Between a Policy Engine and Its Memory Store
- Freshness SLOs: A Staleness Budget per Source
- Invalidation Hooks for Business Data
- Memory Expiry and Decay: Scheduling What an Agent Forgets
- When to Cache Agent Memory: Trade-offs and Strategies
- A Freshness Gate for Agent Memory, Not Just Documents
- All of Memory expiry and staleness