getzep/zep

Zep is a platform that helps AI agents remember and use context across conversations. The core problem it addresses is that AI agents often lack the information they need to give accurate, relevant responses, especially when that information comes from previous conversations, business data, or documents that change over time. Zep collects that information, builds a structured map of the relationships within it, and then retrieves the right pieces at the right moment for the agent to use. The system works in three steps. You feed it data as it arrives: chat messages, business records, events from your application. Zep processes that data automatically, building what it calls a temporal knowledge graph that tracks not just facts but also how those facts have changed over time. When the agent needs context for a new request, Zep assembles a relevant summary quickly, aiming for response times under 200 milliseconds. The technical foundation is a separate open-source project called Graphiti, which is the knowledge graph framework that Zep is built on top of. Graphiti handles the storage of facts with timestamps indicating when each fact was valid, so the system can reason about current versus outdated information. The repository itself is primarily a collection of examples and integration packages rather than the core Zep server, which is a managed cloud service. The examples show how to connect Zep to popular AI frameworks like LangChain, LlamaIndex, and AutoGen. SDKs are available for Python, TypeScript, and Go. The original self-hosted open-source version of Zep, known as Community Edition, has been deprecated and moved to a legacy folder. The project now focuses on the cloud service and the example integrations in this repository.