Definition

What it is

An open-source AI memory platform (Apache 2.0) built around the **Graphiti** temporal-knowledge-graph engine. Zep stores semantic facts as attributes directly on graph edges between entity nodes; every node and edge carries `valid_at` and `invalid_at` properties, letting agents traverse historical states, reason about knowledge decay, and maintain episodic memory autonomously. Repository: [github.com/getzep/graphiti](https://github.com/getzep/graphiti). Release cadence: **107 releases** as of 2026 — one of the most actively maintained AI memory engines.

Why it exists

Static text-as-memory loses the relational structure that gives memory its meaning. Zep transforms text into dynamic triplets (entity nodes + edges representing relationships between them) so that retrieving "what did Sarah work on in Q3 2025" returns not a paragraph but a traversable graph with explicit time-bound edges. This relational + temporal structure dramatically improves recall fidelity on multi-entity reasoning tasks where flat vector search degrades.

Primary use cases

Temporal knowledge graphs for AI agents, multi-agent memory coordination, episodic memory with valid-time semantics, agent reasoning over evolving relationships, S3-backed graph persistence for memory durability.

Recent developments

Latest signals

Zep ranks #1 on LongMemEval with 63.8%. Highest score on the industry-standard long-context memory benchmark — outperforming Mem0, OpenAI Memory, and Anthropic's contextual memory in published comparisons. Per WeavAI — Zep 2026 Review: AI Agent Temporal Memory King.
DMR Benchmark: 94.8% (GPT-4 Turbo) / 98.2% (GPT-4o Mini). Domain-specific Memory Retrieval scores at the top of the table, validating Zep's "graph + temporal" architecture for production memory recall. Per WeavAI Review.
SOC 2 Type 2 + HIPAA + GDPR compliance. Zep has become the shortcut for financial, medical, and legal AI agents to pass enterprise RFPs — compliance certifications are the differentiator vs Mem0 in regulated verticals. Per WeavAI Review.
The original "Zep paper" (arXiv 2501.13956) is the canonical temporal-KG-for-agents reference. Published academic frame: temporal knowledge graphs with bi-temporal modeling (valid time + transaction time) for agent memory. Many 2026 derivative implementations (GraphZep TypeScript port, etc.) trace lineage back to this paper. Per arXiv 2501.13956 — Zep: A Temporal Knowledge Graph Architecture for Agent Memory.
Graphiti is now positioned as Zep's "context graph engine." 2026 Zep messaging frames Graphiti as the open-source temporal context-graph engine "at the core of Zep's context infrastructure" — Zep is the platform; Graphiti is the engine that's adoptable independently. Per GitHub — getzep/graphiti.
Atlan comparison: Zep wins on temporal reasoning; Mem0 wins on ecosystem breadth. Zep's edge is bi-temporal graph queries (what was true in March vs what's true now); Mem0's edge is 21 integration targets and Apache-2.0 simplicity. Pick by workload. Per Atlan — Zep vs Mem0 Benchmarks Pricing.