Every AI tool traps your context in isolated silos. Claude remembers yesterday's conversation. ChatGPT forgets your coding preferences. Cursor starts fresh. This fragmentation forces users to constantly rebuild context when switching between assistants. Open Memory Protocol (OMP) tackles this problem with an open standard for cross-tool memory sharing.

OMP provides a vendor-neutral specification defining memory objects, storage formats, and HTTP APIs. Developers can self-host the reference server using Docker with a single command, while TypeScript and Python SDKs enable integration. The system supports three memory types: episodic events, semantic facts, and procedural knowledge. Claude connects via MCP, and a browser extension bridges web-based AI tools.

The protocol exposes standard REST endpoints for CRUD operations and semantic search. Authentication uses bearer tokens, and memories export as portable JSON. Each memory object includes content, type, source metadata, and tags. The browser extension silently syncs conversations every two minutes, generating natural handoff briefs when continuing discussions in different tools.

OMP represents an attempt to standardize what has been a fragmented frontier. By treating memory as a portable resource rather than tool-specific state, it could reduce the friction users experience when jumping between AI assistants. The project needs adapters for OpenAI, Cursor, Copilot, and Gemini to reach full utility.