Zep provides a memory foundation for AI agents, enabling continuous learning and personalized experiences by storing and retrieving conversational data. It targets AI developers and researchers building sophisticated conversational AI applications, offering enhanced recall and temporal reasoning beyond simple chat history.

How It Works

Zep utilizes a temporal Knowledge Graph, powered by its Graphiti library, to store facts about user interactions. This approach allows Zep to manage state changes and maintain historical context, offering superior retrieval compared to semantic search alone. Facts are precomputed asynchronously for fast retrieval, with speed dependent on the embedding service. Zep supports various data types, including chat messages and structured/unstructured text, and can manage graphs at user, session, and group levels.

Quick Start & Requirements

• Install server: ./zep pull then ./zep up
• Install SDKs: pip install zep-python or npm i @getzep/zep-js
• Prerequisites: OpenAI-compatible LLM API key (e.g., OPENAI_API_KEY environment variable), potentially LiteLLM for other providers.
• Documentation: Quick Start, Documentation, Zep Cloud Docs

Highlighted Details

• Temporal Knowledge Graph for stateful memory and provenance.
• Framework-agnostic: Integrates with LangChain, LangGraph, Chainlit, Autogen, etc.
• Retrieval combines BM25, semantic, and graph search, reranked by distance.
• SDKs available for Python, TypeScript, and Go.

Maintenance & Community

The Zep Community Edition is deprecated and no longer maintained. The project directs users to Zep Cloud. The repository contains a legacy branch for the older Zep Open Source v0.x.

Licensing & Compatibility

The README does not explicitly state a license for the Zep Community Edition or its underlying components. It mentions Zep Cloud is SOC II Type 2 certified.

Limitations & Caveats

Zep Community Edition is deprecated and no longer maintained. There is no migration path from Zep Open Source v0.x to Zep Community Edition. Document Collections and local embedding/NER services are no longer offered.