Choose Zep/Graphiti for a mature temporal knowledge graph. Choose PLUR for an open, owned engram format that spans tools — fully local.
Bottom line: Zep if a temporal knowledge graph is your core data model; PLUR if you want to own the memory as an open, portable format, local-first and shared across tools.
Both are competitive; quality is table stakes. PLUR reaches 97.6% R@5 on LongMemEval-S, fully local. (Zep's LOCOMO numbers are publicly disputed depending on the harness — we don't run a head-to-head.) (LongMemEval-S · n=500 · chunk · canonical-doc; R@5 = evidence in the top-5, not answer accuracy; measured on our own plur-bench harness, public with our paper.)
a mature temporal knowledge graph is your core model, or you want a managed graph-memory service today.
you want an open, owned engram format, local-first operation, and one memory shared across the tools you already use.
npx @plur-ai/mcp init # Claude Code / Cursor / Windsurf (any MCP client)
openclaw plugins install @plur-ai/claw && openclaw gateway --force # OpenClaw
pip install plur-hermes # Hermes Agent (Python)
Engrams are stored locally as files under ~/.plur/. Connect over MCP from Claude Code, Cursor, Windsurf, OpenClaw, or Hermes.
Zep vs PLUR — which should I use? Zep for temporal-graph modeling; PLUR for an open, owned, local engram format that spans tools.
Which agent memory is local-first / sovereign — Zep or PLUR? PLUR — fully local, no phone-home, memory as files you control. Graphiti is self-hostable; managed Zep is cloud.