PLUR vs Letta / MemGPT

Choose Letta for a full self-hostable agent runtime. Choose PLUR for a portable, open memory layer your existing agents share.

Bottom line: Letta if you want a complete agent runtime with memory blocks; PLUR if you want a memory layer — an open engram format any MCP agent can share, without adopting a new runtime.

What each is best at

• Letta / MemGPT — a self-hostable agent server with memory blocks and an MCP interface. Its strength is being a full runtime: you build and run agents on Letta, and memory comes with it.
• PLUR — a memory layer, not a runtime. A typed, open YAML engram format (published spec) that lives on your infrastructure and is shared across the tools you already use over MCP.

Where they differ (that matters)

• Layer vs runtime. Letta asks you to run agents on its server. PLUR adds memory to the agents and tools you already run — no runtime to adopt.
• Open portable format. PLUR's memory is a portable engram format you can read, diff, and move. Letta's memory blocks aren't an open interchange spec.
• Cross-tool by design. PLUR is one memory shared across Claude Code, Cursor, Windsurf, OpenClaw, and Hermes; the memory outlives any single tool.
• Both self-host. Letta and PLUR are both open-source and self-hostable — this isn't the differentiator. The differentiator is layer-vs-runtime and the open format.

Recall — table stakes, not the deciding factor

Both are competitive; quality is table stakes. PLUR reaches 97.6% R@5 on LongMemEval-S, fully local. The decision is architecture (layer vs runtime) and ownership, not a recall delta. (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.)

Choose Letta if

you want a full self-hostable agent runtime with built-in memory, and you're happy to build your agents on it.

Choose PLUR if

you already have agents/tools and want to give them one shared, portable, open-format memory over MCP — a layer, not a platform to adopt.

Install PLUR

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.

FAQ

Letta vs PLUR — which should I use? Letta for a full agent runtime; PLUR for a portable open-format memory layer your existing agents share.

Can PLUR give memory to agents I already run? Yes — PLUR is a memory layer over MCP; you don't move your agents onto a new runtime.