Zep

// Overview

Zep is a context engineering platform that gives AI agents long-term memory. As your app sends it chat messages, business data, documents, and events, Zep extracts the relationships between them and keeps a temporal knowledge graph that tracks how that context changes over time. When the agent needs to respond, Zep returns pre-formatted context blocks built for the LLM.

It is aimed at developers building production AI assistants and agents that need to remember facts about users and how those facts evolve, rather than re-reading the whole conversation on every turn. You add context as events happen and retrieve assembled, relationship-aware context when you call your model.

In the agent-memory category, Zep sits as a memory framework backed by Graphiti, an open-source temporal knowledge graph engine. Each fact carries valid_at and invalid_at dates, so the graph can express when something became true and when it stopped being true. Zep is delivered as a managed cloud service with SDKs for Python, TypeScript, and Go.

// What it does

• Temporal knowledge graph (powered by Graphiti) that tracks facts and relationships with valid_at / invalid_at dates as they change over time
• Graph RAG retrieval that returns relationship-aware context instead of raw chat logs
• Ingests multiple sources: chat history, business data, documents, and app events
• Pre-formatted context blocks built for direct use in your LLM prompt
• SDKs for Python, TypeScript/JavaScript, and Go
• Managed Zep Cloud service with sub-200ms retrieval latency and SOC2 Type 2 / HIPAA compliance

// Getting started

Zep is delivered as a managed cloud service. Sign up for Zep Cloud, then install the SDK for your language and feed context to your agent.

# Python
pip install zep-cloud

# TypeScript / JavaScript
npm install @getzep/zep-cloud

# Go
go get github.com/getzep/zep-go/v2

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Give a chat assistant long-term memory of user facts and preferences across many sessions
• Track how facts change over time (e.g. a user's plan, address, or goals) using the temporal knowledge graph
• Assemble relationship-aware context from chat plus business data and documents before each LLM call
• Add memory to agents built on frameworks like LangChain, LlamaIndex, or AutoGen via the example integrations

// How Zep compares

Zep alongside other open-source agent memory tools AI/TLDR tracks, ranked by GitHub stars.