Zep

Executive Summary

Zep is a context engineering and agent memory platform that uses temporal knowledge graphs to give AI agents persistent, relationship-aware memory. Unlike vector-based RAG approaches, Zep automatically extracts entities, relationships, and facts from conversations and business data, maintaining a graph that evolves over time and invalidates stale information. The company published a peer-reviewed paper demonstrating state-of-the-art performance on agent memory benchmarks. As of June 2026, Zep positions itself as "agent memory, at enterprise scale," anchored by a governed "Context Lake" managing millions of context graphs with access control, retention policies, and audit trails — and claims sub-200ms retrieval (161ms p95 across 10M graphs). An April 2026 S&P Global Market Intelligence report suggested Zep could become "a de facto partner in this layer of the enterprise agent stack."

---

How It Works

Zep operates in three stages: Ingest, Graph, and Assemble.

1. Ingest — Chat messages, JSON business data, and documents are sent to Zep as "episodes." The system automatically extracts entities, relationships, and facts using LLMs.

2. Graph — Extracted data feeds into a temporal knowledge graph. When facts change (e.g., a user moves cities), old edges are invalidated with timestamps rather than deleted, preserving historical context.

3. Assemble — When an agent needs context, Zep retrieves relevant subgraphs using semantic, keyword, and graph-based search, then formats results for token-efficient LLM consumption.

The underlying engine is Graphiti, Zep's open-source temporal knowledge graph framework, which has grown to roughly 27.3K GitHub stars as of June 2026. Graphiti supports Neo4j and FalkorDB as graph stores and works with OpenAI, Azure OpenAI, Google Gemini, and Anthropic models. It also ships an MCP server (v1.x) for integration with Claude, Cursor, and other MCP-compatible clients.

Recent Releases

Graphiti development has been active through 2026: v0.28.2 (March 2026) shipped a security hardening release against Cypher injection in search filters, v0.29.0 (April 2026) brought major efficiency and internal architecture changes, and v0.29.2 (June 8, 2026) added FalkorDB bug fixes.

Benchmark Performance

Zep claims an aggregate accuracy improvement of up to 18.5% over baseline on the LongMemEval benchmark, and outperforms MemGPT (Letta AI) on the Deep Memory Retrieval metric. The benchmark uses conversations averaging ~115,000 tokens. The current site claims 94.7% accuracy on memory benchmarks and 161ms p95 retrieval latency at 10M-graph scale.

Adoption

As of June 2026, Zep's site lists customers and users including Amazon Web Services, Samsung, Twin Health, Praktika.ai, Thrive AI Health, HoneyBook, Harper, FlockX, and a Fortune 500 tech company.

---

Pricing

Pricing moved significantly upmarket in 2026. The previous $25/month Flex tier (20,000 credits) is gone; paid plans are now annual.

Credits are consumed at 1 credit per 350 bytes ingested (or part thereof); webhook invocations cost 1/8 credit, and retrieval, storage, and user management are free. Flex includes 30-day credit rollover, Flex Plus 60-day. Enterprise deployments include managed Cloud, Cloud + BYOK, and BYOC (your VPC) options with SOC 2 Type II and HIPAA BAA compliance.

---

Strengths

• Temporal reasoning — Unlike static vector stores, the graph tracks when facts change and maintains history, enabling "what did we know when" queries
• Published research — Peer-reviewed arXiv paper with reproducible benchmarks gives credibility to performance claims
• Open-source core — Graphiti is Apache 2.0 licensed with ~27.3K stars, usable independently with Neo4j or FalkorDB for self-hosted setups
• Enterprise traction — Named customers including AWS and Samsung, plus SOC 2 Type II / HIPAA compliance and BYOC deployment
• Domain customization — Custom entity types and relationship models adapt to specific business domains
• Framework integrations — Works with LangChain, LlamaIndex, Vercel AI SDK, and offers MCP server support

Cautions

• Cloud-first model — The full Zep platform is cloud-only; the legacy Community Edition is deprecated, so self-hosting is limited to the Graphiti OSS layer without orchestration features
• Rising entry price — The $25/month entry tier was eliminated; the cheapest paid plan is now $1,250/year, a meaningful jump for small teams
• LLM dependency — Graph construction requires LLM calls for entity/relationship extraction, adding latency and cost on top of Zep's credit pricing
• Early performance issues — Reddit users reported initial performance problems in production, though Zep documented a 30x scaling improvement
• Credit-based pricing opacity — Episodes larger than 350 bytes consume multiple credits, making cost prediction harder for high-volume use cases
• Thin funding — Roughly $500K raised versus Mem0's $24M Series A leaves Zep comparatively under-capitalized in a competitive category
• Graph store dependency — Graphiti requires Neo4j or FalkorDB, adding infrastructure complexity for self-hosted deployments

---

What Developers Say

From the Show HN discussion of Graphiti:

"You are definitely onto something here." — tcdent

"I prefer those is because the ontology is already well defined in a lot of cases which is 80% of the battle" — spothedog1 (arguing for predefined ontologies over LLM-extracted ones)

"you're kinda missing the point, there is an existing eco system of ontologies and technologies using RDF" — mehh

The pattern as of June 2026: genuine enthusiasm for the temporal graph approach, paired with skepticism from knowledge-graph veterans about reinventing ontology standards, and recurring complaints in community threads about credit-math pricing and the deprecated self-hosted edition.

---

Competitive Positioning

---

Bottom Line

Zep remains the most technically ambitious approach to agent memory on the market. The temporal knowledge graph architecture is a genuine differentiator — most competitors rely on vector similarity search, which loses relationship context and has no concept of time. Graphiti's growth to ~27K stars validates the open-source strategy, and the 2026 enterprise repositioning (Context Lake, AWS/Samsung logos, S&P Global recognition) shows real commercial momentum despite a tiny ~$500K funding base.

Recommended for: Teams whose use case genuinely needs temporal reasoning about changing facts and relationships, and enterprises wanting governed, compliant agent memory with BYOC deployment.

Not recommended for: Small teams priced out by the $1,250/year entry point, or anyone requiring a fully self-hosted commercial platform — only the Graphiti engine is self-hostable.

Outlook: The move upmarket is deliberate: annual pricing, enterprise governance features, and named Fortune 500 logos suggest Zep is chasing the enterprise agent stack rather than indie developers. The open question is capitalization — better-funded rivals like Mem0 ($24M Series A) can outspend Zep on go-to-market, so expect Zep to lean harder on Graphiti's community gravity and its published research edge.