Plur

PLUR is an open standard and toolkit for persistent AI agent memory. It gives coding agents one shared memory that carries across tools, models and sessions, so a correction made in Claude Code is already known to Cursor, and a pattern learned by a subagent is available to an orchestrator on its next run. Memory is stored as engrams, which are single typed units of learned knowledge such as a correction, a preference or a fact, written as plain YAML in a local directory. Users can open, edit or delete any engram and the change applies immediately, with no retraining. The engram format is an open specification under the Apache 2.0 license, first published in March 2026. Implementations include a TypeScript core, an MCP server, plugins for DeepSeek Harness and OpenClaw, and a Python SDK, installed through npm and started with an init command. The vendor reports 97.6 percent recall at five on the LongMemEval benchmark with fully local retrieval, and a task win rate for agents with memory against agents without it. PLUR suits developers and engineering teams who run several coding agents and want memory they own and can inspect. A separate enterprise offering runs the open core self hosted in a customer's own VPC or hardware, with OIDC, SAML 2.0 and GitHub OAuth sign in, SCIM provisioning, role based access, audit logs and a REST API. The open source software is free. Enterprise pricing is not published, and the vendor directs interested teams to talk to them.