MemClaw vs. Zep

Zep is a capable memory layer with an auto-extracted knowledge graph and PII handling — strong for a single agent or assistant. The two overlap on graph and PII, then diverge on fleet shape: MemClaw governs memory shared across many agents, teams, and vendors, where Zep is built around single-agent recall.

CapabilityMemClawZep
Multi-fleet support
Agent trust tiers + keystone policies
Cross-vendor memory sharing
Contradiction detection + supersession
Per-agent retrieval tuning
PII detection & quarantine
Audit trail / provenancePartial
Knowledge graph (auto-extracted)
MCP-native
OSS licenseApache 2.0Apache 2.0

Where MemClaw differs

  • Multi-fleet, multi-tenant by design. Memory is scoped and shared across fleets and vendors on one plane, not bound to a single agent’s session.
  • Trust tiers + keystone policies. Per-agent trust levels and mandatory keystones govern what each agent can read and write — enforced server-side, not by prompt convention.
  • Contradiction detection + supersession. Conflicting facts are detected and the stale one is superseded, with a full audit trail of every transition.
Start freeRead the docsView on GitHub

Comparison reflects our reading of each project’s public documentation as of June 2026. Mem0, Zep, and Letta are solid projects for single-agent memory. Spot something out of date? Open an issue or PR and we’ll correct it.