Hypermemory vs Zep
Both give AI agents long-term memory. The difference is how they structure it, and what happens when you grow.
| Aspect | Hypermemory | Zep |
|---|---|---|
| Architecture | Hypergraph (multi-way relationships) | Temporal Knowledge Graph (binary edges) |
| Billing Unit | Nodes and queries (intuitive) | Episodes / credits (variable) |
| Credit Rollover | N/A (no credits) | 60-day rollover |
| Enterprise | Custom pricing | Custom pricing |
| Entry Paid Tier | $15/mo (15K queries + nodes) | $25/mo (20K credits) |
| Free Tier | 6,000 queries + nodes (one time) | 1,000 credits |
| Large Data Penalty | None | Episodes >350 bytes billed as multiples |
| MCP Support | Native (built for MCP) | Available (add-on) |
| Mid Tier | $30/mo (50K queries + nodes) | $475/mo (300K credits) |
| Overage Pricing | $0.61/1K queries + nodes (every paid plan) | Must upgrade tier (cliff) |
| Price Jump | 2x ($15 → $30) | 19x ($25 → $475) |
| Pricing Model | Nodes + Queries (transparent) | Credits / Episodes (opaque) |
| Relationship Model | Hyperedges (3+ nodes per relationship) | Binary edges (A → B only) |
| SSO/SAML | Enterprise tier | Enterprise tier |
| Storage Fees | None | None |
What happens when you grow
Zep's pricing has a cliff. Hypermemory's doesn't.
| Monthly Usage | Zep Cost | Hypermemory Cost | Difference |
|---|---|---|---|
| 20K operations | $25 | $18.05 (Basic + overage) | Save $6.95 |
| 25K+ operations | $475 (must upgrade) | $21.10 (Basic + overage) | Save $453.90 |
| 50K operations | $475 | $30 (Pro) | Save $445 |
| 100K operations | $475 | $60.50 (Pro + overage) | Save $414.50 |
| 500K operations | $475+ | $304.50 (Pro + overage) | Save $170.50+ |
Binary edges vs. hyperedges
Zep uses a Temporal Knowledge Graph, a powerful upgrade over pure vector search, and a legitimate approach to structured memory. But its relationships are binary: A connects to B, and that's it.
Hypermemory uses a hypergraph. A single relationship can connect three, four, or more concepts simultaneously. When your agent remembers that "Sarah, Marcus, and the API v2 project were all part of the Q3 planning decision", that's one hyperedge, not six separate binary edges. The same graph is available from every tool you connect.
This matters because real knowledge is rarely just two things connected. Decisions involve multiple people. Projects span multiple teams. Context is multi-dimensional. Hyperedges capture that naturally.
Honest guidance
Choose Zep if:
- You're already deeply integrated with Zep's ecosystem
- Your usage stays comfortably under 20K episodes/month
- You need Zep's specific GraphRAG capabilities
- You're on an enterprise plan with negotiated pricing
Choose Hypermemory if:
- You want lower prices from the start, $15/mo vs. Zep's $25/mo
- You need multi-way relationships (hyperedges), not just binary connections
- You want MCP-native integration with no adapters
- You want transparent, predictable pricing that scales smoothly
- You want 50K queries + nodes for $30/mo instead of paying $475 for 300K credits
Competitor pricing last verified: 2026-04-30