The architectural difference
Zep and SolarflareDB overlap around helping applications recover useful context, but they place the product boundary in different locations. The right choice depends on whether the application needs a specialized retrieval or memory service, a general graph engine, or a unified authoritative temporal context layer.
Capability matrix
| Dimension | Zep | SolarflareDB private-beta direction |
|---|---|---|
| Primary abstraction | Temporal agent context platform | Temporal context database |
| Graph construction | Managed temporal knowledge graph | Episode/fact/entity graph with optional extraction boundary |
| Freshness contract | Validate current ingestion-to-query behavior | Write bookmark + recent authoritative overlay + index watermarks target |
| Deployment | Review current hosted options | Cloudflare-managed target, future Core and BYOC |
| Evaluation | Use vendor docs and own benchmark | Open workload, cost, temporal, entity, and provenance harness target |
| Pricing position | Credit/subscription model; review current page | $0/$5 subsidized entry and action-based overage proposal |
When to choose Zep
Choose Zep when you want an established agent-memory and temporal context product and its current API, hosting model, retrieval behavior, and commercial plan fit the workload.
Retaining a focused system is usually better than introducing a new database merely because it has a broader feature list. Benchmark the actual workload and operational boundary.
When to choose SolarflareDB
Choose SolarflareDB when you want a lower subsidized entry tier, transparent operation meters, explicit authoritative/derived watermarks, portable object-store recovery, and typed merge semantics.
SolarflareDB should earn adoption by reducing custom context infrastructure and improving temporal correctness, retrieval explainability, and cost transparency—not by claiming every incumbent is obsolete.
How to run a fair evaluation
- Use the same source episodes, identity rules, embedding model, and extraction policy.
- Measure immediate and settled retrieval after writes.
- Include current truth, historical truth, entity aliases, contradictions, and multi-hop questions.
- Record context tokens, answer quality, latency, index freshness, and operations cost.
- Test deletion, export, duplicate delivery, and concurrent writers.