PLUR vs claude-mem

Choose claude-mem for lightweight session context with zero config. Choose PLUR for structured knowledge — engrams that accumulate, decay when stale, and transfer via the pack exchange layer.

Bottom line: claude-mem if you want fast session replay — compressed observations injected into the next session; PLUR if you want memory that builds knowledge — structured facts that improve with use, forget intelligently, and share across agents and teams.

What each is best at

• claude-mem — hooks into sessions and compresses tool calls and outputs (~500 tokens per observation) for injection into future sessions. Lightweight, widely adopted, zero configuration. Strength: fast, session-aware context from what the agent did.
• PLUR — stores engrams: atomic facts the agent concluded, with confidence scores, ACT-R activation decay, and a shared exchange protocol. Strength: knowledge that accumulates, improves with feedback, and transfers via the pack exchange layer — not a log, a knowledge base.

Where they differ (that matters)

• Storage unit — observation vs engram. claude-mem stores: "ran npm install express in session 2026-07-19." PLUR stores: "use pnpm, not npm — workspace:* dependencies break on npm install." One is a record; the other is a reusable principle that fires whenever it's relevant.
• Exchange layer — silo vs protocol. claude-mem memory lives in a single agent. PLUR's exchange protocol lets teams bundle engrams as knowledge packs, install community packs, and share memory across agents and tools via ~/.plur/. No other memory system has a pack exchange layer.
• Memory decay — accumulate vs forget intelligently. claude-mem accumulates observations. PLUR applies ACT-R activation decay: engrams that aren't reinforced by recall lose weight over time. Stale facts are deprioritized, not inherited forever. Memory that forgets the right things is more useful than memory that grows without bound.

What they share

Both claude-mem and PLUR default to fully local — no cloud, no API key, no data leaving your machine. Both work with Claude Code and MCP-compatible clients. Both are open source (Apache-2.0). The market they're validating together is real: local, agentic memory for AI developers.

Choose claude-mem if

you want zero-configuration session context — it installs fast and injects what the agent did in previous sessions without any setup or structured memory model.

Choose PLUR if

you need memory that builds durable knowledge rather than a session log, decays stale facts intelligently, transfers across agents via the pack exchange layer, works across Cursor and Windsurf in addition to Claude Code and Hermes, or persists team-level conventions that outlive any single session.

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/ — your files, your infrastructure. Connect over MCP from Claude Code, Cursor, Windsurf, OpenClaw, or Hermes.

FAQ

claude-mem vs PLUR — which should I use? claude-mem for lightweight session context with zero config; PLUR for structured knowledge that accumulates, improves, and transfers across agents and tools.

What is the difference between claude-mem and PLUR? claude-mem captures compressed session observations — what the agent did (tool calls, outputs). PLUR stores engrams — atomic facts the agent concluded. A session log vs a knowledge base. Both default to fully local with no cloud dependency.