Memtro vs Zep
A developer memory platform with a temporal graph, versus an organisation-wide memory with connectors, routing and jobs.
This comparison describes Zep as we understand it at the time of writing from its public documentation. Products change; check the current details before deciding.
Zep is a memory platform for AI agents built around a temporal knowledge graph (Graphiti). You add it to your agent code to store conversations and business data and retrieve relevant context, with cloud and self-hosted options.
Memtro also keeps a knowledge graph with time-aware facts, but packages it as an OpenAI-compatible endpoint and MCP server with connectors, routing, agent mode and jobs, so existing tools use it without code.
Where they differ
| Memtro | Zep | |
|---|---|---|
| Shape | Endpoint and MCP server in front of your tools and models | SDK and API inside your agent application |
| Graph | Nodes, edges and typed memories with validity dates and supersession, in Postgres with pgvector | Temporal knowledge graph (Graphiti) |
| Data in | Conversations, connector tool results, hourly syncs, imports from ChatGPT and Claude | What your application sends |
| Connectors | Fifteen read-only connectors; results anchored deterministically | Bring your own data |
| Personal vs shared | Classified per fact under the request's scope | Per user and group, as configured by your app |
| Agent runtime | Built in: agent mode, OpenCode and Claude Code engines, jobs | Your agent framework |
| Hosting | Hosted platform | Cloud or self-hosted |
When to choose which
Choose Zep when you are building an agent and want a rich, developer-controlled memory graph with your own retrieval logic. Choose Memtro when the goal is a company-wide memory that the tools people already use can share, with the connectors, routing and scheduling included.