Open Memory Protocol Initiative

The Open Memory Protocol (OMP) is an open protocol for moving and sharing AI-agent memory. Agent memory is persistent context that affects future model or agent behavior: user-provided facts, model- or agent-derived summaries, project instructions, prior decisions and actions, tool outputs, and references to underlying transcripts or artifacts.

Every major coding-agent harness (Claude Code, Codex, goose, OpenClaw, Letta Code) implements memory, but each uses a different convention: AGENTS.md, MEMORY.md, USER.md, HUMAN.md. Developers cannot move a stateful agent from one harness to another. Enterprises cannot switch memory providers without re-ingesting derived memory objects.

The fragmentation has real engineering and security costs. Provenance is often lost when memory is copied, and permissions that were meaningful in one system may not survive export. OMP is the smallest interoperable technology layer that addresses these costs, while allowing implementers to keep their own storage backends, ownership models, and internal architectures.

The first implementation focus is coding agents, where memory lock-in is most acute. Early implementations explore memory transfer across Claude Code, Codex, goose, and Letta Code.