Research-based 7-layer memory vs. fast chat memory with automatic fact extraction.
March 2026
15 algorithms (9 + 6 PMA), Sleep Consolidation, Retention Curves
Fast chat memory with fact extraction and session summaries
Zep is a strong chat memory system with fact extraction and session summaries — ideal for chatbot developers who need fast memory. ZenBrain goes significantly deeper: 7 memory layers, neuroscience algorithms like Sleep Consolidation and Hebbian Learning, and a MemoryCoordinator that orchestrates all layers. For projects that go beyond chat memory, ZenBrain is the better choice.
| Feature | ZenBrain | Zep |
|---|---|---|
| Memory Layers | 7 layers | 2 (Facts + Sessions) |
| Sleep Consolidation | ✓ | ✗ |
| Fact Extraction | Entity Resolution | ✓ |
| Session Summaries | Episodic Memory | ✓ |
| Spaced Repetition | ✓ | ✗ |
| Temporal Awareness | ~ | ✓ |
| Self-Hosted | ✓ | ~ |
| Zero Dependencies | ✓ | ✗ |
ZenBrain is a fully open-source, self-hostable memory library with seven neuroscience-grounded memory layers and algorithms such as FSRS spaced repetition, Hebbian learning and Ebbinghaus decay. On Zep’s side the open part has shifted: per Zep’s own README the Community Edition is no longer supported and its code sits in a legacy/ folder. What remains open and actively maintained is Graphiti, the Apache 2.0 temporal knowledge-graph framework that powers Zep Cloud; the getzep/zep repository is explicitly not the product but holds examples, integrations and benchmarks. ZenBrain targets developers who want full control of the memory architecture; Zep targets teams that prefer a managed memory API.
ZenBrain (@zensation/algorithms and @zensation/core) is Apache 2.0 and free to use, including commercially — no usage fees, no rate limits, no per-user pricing. On Zep’s side Graphiti is likewise Apache 2.0 and free; what is commercial is the hosted Zep Cloud platform. The former Community Edition is no longer a point of difference, because it has been discontinued.
No — and that is the honest position. Temporal reasoning is roadmap for ZenBrain and not wired: the architecture carries timestamps and an episodic memory, but we have no benchmark result on the temporal axis; an internal run scored zero per cent there because the timestamp was not passed through. If you need temporal reasoning over a knowledge graph, Graphiti is the better fit — it is open source under Apache 2.0 and built for exactly that.
ZenBrain includes a sleep-consolidation module that simulates offline replay: it selects important memories for reinforcement and prunes weak associations. The mechanism is conceptually grounded in sleep-dependent memory consolidation research (e.g. Stickgold & Walker, 2013, https://doi.org/10.1038/nn.3303). A comparable mechanism is not described in Zep’s or Mem0’s public documentation as of 16 August 2026 — a statement about their documentation, not about their implementation.
Yes. The MemoryCoordinator orchestrates all seven memory layers through a unified API: store(), recall(), consolidate(), decay() and FSRS review queues. The same algorithms underpin ZenAI. Zep is production-ready too; per the vendor its most capable features run primarily through the hosted Zep Cloud service, whereas ZenBrain runs entirely on infrastructure you control.