Zep

Zep is a context engineering platform for AI agents built on Graphiti, an open-source temporal knowledge graph engine with 31.2k GitHub stars. Unlike traditional memory systems that rely on static RAG or chat history alone, Zep automatically extracts entities, relationships, and facts from conversations and business data. It then builds a unified context graph that evolves as information changes. When a fact is superseded (for example, when a user changes their preferences or a business relationship shifts), Zep automatically invalidates the old fact while preserving its historical context and provenance. This temporal awareness distinguishes Zep from memory layers built around a vector store, such as Mem0. The core workflow has three steps. Ingest: chat messages, JSON business data, and documents sent via the API. Build: Graphiti constructs a temporal knowledge graph with automatic entity extraction and fact invalidation. Assemble: Zep retrieves relevant context and formats it for the LLM at under 200ms P95 retrieval latency. Named customers include Samsung, Writer, HoneyBook, Twin Health, Thrive AI Health, Praktika.ai, AGI Inc, Harper, FlockX, and Aurasell. On the LoCoMo benchmark, Zep publishes 94.7% accuracy at 155ms with single-shot retrieval and no slow agentic loops. On LongMemEval it publishes 90.2% accuracy at 162ms. Both figures are published by the vendor from its own measurements on two industry benchmarks, restated here rather than independently reproduced. The Graphiti MCP Server gives Claude Desktop, Cursor, VS Code with Copilot, and other MCP-compatible clients persistent knowledge graph memory across sessions. A second, managed Memory MCP Server became generally available on all plans on 19 August 2026, on Zep managed cloud and BYOC, binding each person to their own user graph through Google Workspace or an enterprise identity provider and metering access in per-account MCP seats. Python, TypeScript, and Go SDKs are available. Works with LangChain, LangGraph, CrewAI, and any agent framework via API-first architecture with webhook support. Named gaps: no no-code interface for non-developers. Self-hosted Graphiti requires Docker, FalkorDB or Neo4j, and LLM API key management. HIPAA BAA is available on Enterprise only. Pricing verified live on getzep.com/pricing as of July 2026. Free tier: 10,000 credits per month for prototyping with limits applied. Flex at $125 per month ($104 per month billed annually at $1,250 per year, saving 17%) includes 50,000 credits per month, auto-topup at 20% with 30-day rollover, 600 requests per minute, 5 projects, and 10 custom entity and edge types. Flex Plus at $375 per month ($312.50 per month billed annually at $3,750 per year) includes 200,000 credits per month, 60-day rollover, 1,000 requests per minute, 10 projects, 20 custom entity and edge types, custom extraction instructions, webhooks, and analytics. Enterprise is custom-priced with on-premises or BYOC deployment, SSO, HIPAA BAA, dedicated SLA, VPC deployment, and 1-year audit logs. Overage on Flex is $25 per 10,000 credits; on Flex Plus is $75 per 40,000 credits. Credits are consumed based on Episode size: 1 credit per Episode up to 350 bytes, plus 1 credit per additional 350 bytes. Retrieval, storage, threads, users, and graph storage consume zero credits. Zep is not the right fit for teams that need a no-code memory interface. The platform is developer-facing and requires API or SDK integration. Teams on tight budgets face a pricing gap: the free tier is limited to 10,000 credits per month for prototyping, then the next tier jumps to $125 per month ($104 annual) with no intermediate option. Teams needing real-time browser or screen context rather than conversation and document memory should evaluate Browser Use (free open-source) instead. As of Q3 2026, Zep carries 31.2k GitHub stars on the Graphiti repository and positions as the context lake for AI agents with enterprise-scale deployment options including cloud, BYOK, and BYOC. Named customers include Samsung, Twin Health, Thrive AI Health, Writer, and HoneyBook. S&P Global Market Intelligence has published coverage of Zep. The platform has launched dedicated comparison pages against Mem0, Letta, AWS AgentCore, Vertex AI Memory Bank, Cognee, and Supermemory. Security: SOC 2 Type II certified via annual third-party audits with continuous monitoring. HIPAA-compliant infrastructure with BAAs available for healthcare organizations. No G2 listing exists as of July 2026.