Zep is a managed platform for building agent memory that provides examples, integrations, and tools for connecting to Zep Cloud.
Zep addresses the challenge of maintaining persistent, contextual memory for AI agents across conversations and interactions. The platform uses a temporal knowledge graph approach to store and retrieve agent memories, allowing agents to build understanding over time rather than treating each conversation in isolation. Developers integrate Zep Cloud into their agent applications through official SDKs available in Python, TypeScript, and Go, then use this repository's examples and framework integrations to connect their specific agent framework to Zep's memory backend.
Teams building agents with popular frameworks should evaluate Zep if they need managed memory that persists across sessions and can be queried semantically. The repository provides integration packages for Python frameworks including LangGraph, CrewAI, Pydantic AI, Microsoft AutoGen, and others, as well as TypeScript support for Google ADK and Vercel AI SDK. The tool suits projects where agents need to accumulate knowledge about users or domains over time, rather than starting fresh with each interaction. Zep Cloud is a hosted service requiring signup; this repository contains only the integration code and examples, not the core product itself.
Development activity shows active maintenance of multiple integration packages across different frameworks and languages, with each integration built, tested, and released independently. The project maintains example code in Python, TypeScript, and Go alongside a bulk data ingestion pipeline supporting Slack, documents, email, and structured data formats. Evaluation tooling is present in the form of memory benchmarks and an ingestion evaluation harness. The team has deprecated the Community Edition and moved its code to a legacy folder, signaling a shift toward the managed cloud service model.
