Engram vs Zep
Compare Engram and Zep for AI agent memory — zero-knowledge encryption, compression, and developer experience.
Overview
Zep is a memory layer for AI assistants focused on conversation summarization and knowledge extraction. Engram is an encrypted, compressed memory system with zero-knowledge search.
The key difference: Zep encrypts at rest (the server can still read your data). Engram encrypts client-side — the server searches encrypted vectors without ever seeing plaintext.
Feature Comparison
| Feature | Engram | Zep |
|---|---|---|
| Zero-knowledge encrypted search | ✅ CAPRISE | ❌ (at-rest only) |
| Client-side encryption | ✅ | ⚠️ Partial |
| 10-50× compression | ✅ | ❌ |
| Code memory | ✅ | ⚠️ Limited |
| Token savings | ✅ 50-80% | ⚠️ Partial |
| MCP protocol | ✅ | ❌ |
| Open-source | ✅ Client | ✅ Server |
| Self-host | ✅ Enterprise | ✅ |
Encryption: At-Rest vs Zero-Knowledge
Zep’s encryption at rest means your data is encrypted on disk — but the Zep server decrypts it to process queries. An attacker who compromises the server gets plaintext access.
Engram’s CAPRISE encryption means the server never has plaintext. It searches distance-preserved encrypted vectors directly. Even a full server compromise reveals nothing.
Self-Hosting
Zep offers self-hosting of its full server. Engram’s client is open source, and the Enterprise plan includes VPC/private deployment. For teams that want maximum control, both options exist — but Engram adds zero-knowledge on top.
Bottom Line
Choose Zep if you want self-hostable conversation summarization and don’t need true zero-knowledge encryption.
Choose Engram if you need privacy guarantees that survive a server breach, plus code memory and token optimization.