Overview
Zep is a memory layer for AI agents: it ingests conversations and builds a temporal knowledge graph, so the agent recalls relevant facts — and how they change over time — instead of replaying raw history.
It runs as managed Zep Cloud or self-hosted, and exposes a simple add/get memory API.
flowchart LR msgs[conversation] --> z[(Zep)] z --> graph[temporal knowledge graph] q[next turn] --> z z --> ctx[relevant context]
The Code samples tab shows the add-and-recall flow.
When to use it
Choose Zep when long-term, evolving memory matters — personal assistants and support agents that must remember user facts across sessions and keep them current, beyond a single context window.