Xtended vs Zep: Relational Tables vs Temporal Graphs

The Bottom Line

Choose Zep/Graphiti if: You need temporal reasoning (when facts were valid), sophisticated entity extraction, or are building enterprise agents that need state-of-the-art memory performance.

Choose Xtended if: You want simpler relational queries, prefer familiar data structures, want explicit field selection and auto-expand, or you're working across multiple AI platforms.

What Zep Does Brilliantly

Zep's Graphiti is genuinely impressive technology:

State-of-the-Art Performance

On the DMR benchmark: 94.8% vs 93.4% (MemGPT). On LongMemEval: up to 18.5% accuracy improvement with 90% latency reduction. These aren't marketing numbers—they're published research.

Temporal Awareness

Graphiti's bi-temporal model tracks both when an event occurred AND when it was ingested. Every edge has explicit validity intervals. This enables:

• "What did we know about this customer in March?"
• "When did this fact become obsolete?"
• Point-in-time queries across your knowledge

Automatic Entity Extraction

Feed Zep conversations or documents, and it extracts entities and relationships automatically. The knowledge graph builds itself.

Hybrid Retrieval

Combines semantic embeddings, BM25 keyword search, and graph traversal. P95 latency of 300ms without LLM calls during retrieval.

The Architecture Difference

Zep: Temporal Knowledge Graph

Nodes: Entities (People, Companies, Projects)
Edges: Relationships with time validity
       ├── valid_from: when relationship started
       ├── valid_to: when relationship ended
       └── ingested_at: when we learned this

Query: Graph traversal + semantic search
       "Find entities connected to Project X
        that were active in Q3 2025"

Xtended: Relational Tables

Tables: Structured schemas (Deals, People, Companies)
Rows: Records with explicit fields
Relations: Foreign keys and typed relationships

Query: Relational
       GET /records?template=deals&company_id=123
       &created_after=2025-07-01&created_before=2025-09-30

The Comparison

The Simplicity Trade-off

Knowledge graphs are powerful. They're also complex.

Zep's Mental Model

To use Zep effectively, you need to understand:

• Nodes, edges, and graph traversal
• Bi-temporal validity (occurrence time vs ingestion time)
• Episode-based data ingestion
• How semantic search combines with graph structure

Xtended's Mental Model

Templates, records, relationships. If you've used a database or spreadsheet:

• You already know how to think about it
• REST APIs are familiar to most developers
• Debugging is straightforward

When Temporal Matters

Zep's bi-temporal model shines when:

Facts Change Over Time

// John was CEO from 2020-2023, then became advisor
// Traditional DB: You'd overwrite or add a new row
// Graphiti: The edge "John --[role]--> CEO" has validity [2020, 2023]
//           A new edge "John --[role]--> Advisor" has validity [2023, present]

Query: "What was John's role in 2022?" → CEO
Query: "What is John's role now?" → Advisor

You Need Historical Context

"What did we know about this customer before the acquisition?" Temporal graphs answer this naturally.

Regulatory/Audit Requirements

When you need to prove what you knew and when you knew it, bi-temporal data is invaluable.

When Relational Simplicity Wins

Xtended's approach shines when:

Structured Queries with Auto-Expand

// Get deals with related company and owner info
GET /records?template=deals&expand=company,owner

// Filter by stage and value
GET /records?template=deals&stage=negotiation&value_gt=100000

// Select specific fields
GET /records?template=deals&fields=name,value,company.name

These queries are natural in REST APIs, require graph traversal knowledge otherwise.

Familiar Data Structures

No new query paradigm to learn. No graph concepts to internalize. Ship faster.

Cross-Platform Access

Same knowledge in Claude, ChatGPT, Cursor, your apps. Zep is powerful but primarily serves its own platform.

When to Use Which

Use Zep when:

• Temporal reasoning is core to your use case
• You need best-in-class benchmark performance
• Automatic entity extraction is valuable
• Your team is comfortable with graph concepts
• Enterprise compliance (SOC 2, HIPAA) is required now

Use Xtended when:

• Relational simplicity matters
• Aggregations and structured insights are important
• You need knowledge in multiple AI platforms
• Non-developers need to interact with the system
• You prefer explicit schema control over automatic extraction

Consider both when:

• Zep for complex temporal reasoning about entities
• Xtended for structured data, analytics, and cross-platform access

The Honest Take

Zep/Graphiti is sophisticated technology. The temporal knowledge graph approach, the benchmark performance, the enterprise features—it's genuinely state-of-the-art for its category. If you need what graphs do well, Zep does it better than most.

But graphs aren't always the right tool. When your questions are "show me all deals in negotiation" and "expand the related contacts" rather than "how are these connected over time," relational simplicity is a feature, not a limitation.

Choose based on your actual queries, not architectural elegance. The best tool is the one that answers your questions fastest.