Memory System Design

Memory Architecture Spectrum

Memory System Performance

Why Vector Stores Fall Short

• Can retrieve "Customer X purchased Product Y"
• Cannot answer "What did customers who bought Y also buy?"
• Cannot distinguish current vs outdated facts

Memory Implementation Patterns

Pattern 1: File-System-as-Memory

# Simple, no infrastructure needed
def store_fact(entity_id, fact):
    path = f"memory/{entity_id}.json"
    facts = load_json(path, default=[])
    facts.append({"fact": fact, "timestamp": now()})
    save_json(path, facts)

Pattern 2: Vector RAG with Metadata

# Embed facts with rich metadata
vector_store.add(
    embedding=embed(fact),
    metadata={
        "entity_id": entity_id,
        "valid_from": now(),
        "source": "conversation",
        "confidence": 0.95
    }
)

Pattern 3: Knowledge Graph

# Preserve relationships
graph.create_relationship(
    from_entity="Customer_123",
    relationship="PURCHASED",
    to_entity="Product_456",
    properties={"date": "2024-01-15", "quantity": 2}
)

Pattern 4: Temporal Knowledge Graph

# Time-travel queries
def query_address_at_time(user_id, query_time):
    return graph.query("""
        MATCH (user)-[r:LIVES_AT]->(address)
        WHERE user.id = $user_id
        AND r.valid_from <= $query_time
        AND (r.valid_until IS NULL OR r.valid_until > $query_time)
        RETURN address
    """, {"user_id": user_id, "query_time": query_time})

Entity Memory

• Entity Identity: "John Doe" in one conversation = same person in another
• Entity Properties: Facts discovered about entities over time
• Entity Relationships: Relationships discovered between entities

def remember_entity(entity_id, properties):
    memory.store({
        "type": "entity",
        "id": entity_id,
        "properties": properties,
        "last_updated": now()
    })

Memory Consolidation

• Memory accumulates significantly
• Retrieval returns too many outdated results
• Periodically on schedule
• Explicit request

1. Identify outdated facts
2. Merge related facts
3. Update validity periods
4. Archive/delete obsolete facts
5. Rebuild indexes

Choosing Memory Architecture

Best Practices

1. Match architecture to query requirements
2. Implement progressive disclosure for access
3. Use temporal validity to prevent conflicts
4. Consolidate periodically
5. Design for retrieval failures gracefully
6. Consider privacy implications
7. Implement backup and recovery
8. Monitor growth and performance