claude-mem vs Zep
Zep: A memory server building a temporal knowledge graph from chats.
Zep is a memory server that builds a temporal knowledge graph from chat history — you run it and integrate it into your stack. cmem takes a different path: it installs into your coding agent and captures structured observations automatically as you work, then exposes them to any MCP client. The engine is open source and local-first; CMEM Cloud adds one private sync link.
Move without ripping anything out
Install the engine
Run npx claude-mem@latest install. The open-source engine wires into Claude Code natively and starts writing observations — nothing to migrate first.
Keep what you have
cmem speaks MCP, so it sits alongside Zep rather than replacing it. Point any MCP client at your private link and both can read.
Recall everywhere
Once CMEM Cloud mirrors your DB, the same memory is reachable from every agent and machine through one private MCP link.
Memory that compounds, built in
Automatic capture
A second model writes structured notes out-of-band — no prompting, no slowdown.
Auto-categorization
Decision, bugfix, feature, discovery, security — every note lands in a taxonomy.
Search three ways
Semantic, keyword, and timeline search over everything your agents have done.
Causality threading
Before/after links connect why a change happened to what came next.
Local-first & private
Memory lives in local SQLite with privacy stripping; cloud sync is opt-in.
Smart-explore
Structural codebase exploration that is 6–12× cheaper than reading files in full.