plur-ai/plur

What it solves

PLUR addresses the "amnesia" problem in AI agents, where they fail to remember user preferences, coding styles, or project-specific conventions across different sessions, tools, or machines. It provides a persistent, local-first memory layer that ensures agents learn from your corrections and maintain consistency over time.

How it works

PLUR stores knowledge as "engrams"—plain-text YAML files on your local disk. It uses a hybrid search approach (BM25 keyword matching combined with local BGE embeddings) to retrieve relevant information without needing cloud API calls. As you interact with agents, PLUR automatically captures learnings, manages their relevance through an activation decay model (similar to human memory), and injects the most pertinent context into your agent's prompts.

Who it’s for

It is designed for developers and power users who rely on AI coding assistants (like Claude Code, Cursor, or Codex) and want to maintain data sovereignty over their agent's "learned" knowledge without relying on proprietary cloud-based memory solutions.

Highlights

• Local-first and private: All memory is stored as plain YAML files on your machine; no data is sent to the cloud.
• Cross-tool compatibility: Works across multiple agents and IDEs, including Claude Code, Cursor, Codex, and OpenCode, via the Model Context Protocol (MCP).
• Performance-focused: Benchmarked to outperform larger models (e.g., Haiku + PLUR vs. Opus without memory) at a fraction of the cost by providing better context.
• Open format: Uses a versioned, schema-validated format that you can inspect, edit, or delete at any time, ensuring you are never locked into a black-box system.