PLUR launches persistent memory system for AI agents, storing corrections and preferences locally without cloud calls or costs.

PLUR is a memory system for AI agents (autonomous tools like Claude Code and Cursor) that persists corrections, preferences, and conventions across sessions and machines as plain YAML files stored locally. Installation is a single command; the system automatically captures learnings from agent conversations via plugins for OpenClaw, Hermes Agent, and other tools.

Memory is stored as two primitives: engrams (learned knowledge with activation decay, feedback signals, scope, and polarity) and episodes (timestamped event records). Search uses local BM25 with IDF weighting and BGE embeddings with zero API calls. The system injects relevant engrams into agent context before each task within a token budget.

In benchmarks across 19 contests on Claude Haiku, Sonnet, and Opus, agents with PLUR achieved an 89% win rate (31 wins, 4 losses). Haiku with PLUR scored 0.80 on discoverability versus Opus alone at 0.31. The body reports: 'Haiku with PLUR memory outperforms Opus without it — 2.6x better on tool routing, at roughly 10x less cost.'