PLUR: Persistent memory for AI agents. Local-first, zero-cost (github.com)

PLUR, a new technology for AI agents, offers persistent memory that is local-first and incurs zero costs, significantly enhancing the performance of AI-driven coding tools. Unlike traditional systems that rely on cloud storage or external APIs, PLUR stores corrections, preferences, and conventions as plain YAML files directly on users' devices. This innovation enables seamless continuation of workflows across different sessions and tools, as agents can retain and apply knowledge like coding style adjustments or architecture descriptions without needing fresh input each time. The implications for the AI/ML community are profound, as PLUR has demonstrated exceptional efficacy in tool routing, outperforming its counterpart Opus by 2.6 times while operating at roughly one-tenth the cost. With features like engram creation, feedback systems, and episode tracking, PLUR not only retains the learned knowledge but also adapts and improves based on user interactions. By employing a hybrid search method combining BM25 and embeddings, PLUR retrieves and injects relevant memories with zero API calls, showcasing how next-generation memory systems can dramatically advance AI operational efficiency. Developers can easily integrate PLUR using a single command and enjoy a powerful memory-enhanced AI experience across various applications.