Open Memory Protocol Unifies AI Memory Across Tools
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Open Memory Protocol is positioned as a vendor-neutral specification for how AI tools store, retrieve, and share user memory, framed by its creators as the solution to "the AI memory silo problem" where Claude, ChatGPT, Cursor, and Copilot each remember users only within their own walls.
- OMP ships four components: a formal spec for memory objects and HTTP API, a self-hostable reference server runnable via
npx omp-serveror Docker on port 3456, TypeScript and Python SDKs, and ready-made adapters for Claude (via MCP), OpenAI, and Cursor. - The memory schema defines three types — episodic (events), semantic (facts and preferences), and procedural (how-to knowledge) — with canonical fields for content, source tool, session ID, timestamps, tags, and optional embeddings stored in SQLite or pgvector.
- OMP's portability features include REST CRUD plus semantic-search endpoints, bearer-token authentication with per-tool API keys, and a portable JSON export/import format designed so memories can be moved between servers or pasted into any AI that reads JSON.
- The protocol's roadmap runs from v0.1 (core spec and MCP adapter, currently shipping) through v0.3 (pgvector semantic search), v0.4 (per-project namespacing), v0.5 (multi-user support and access control), to a planned v1.0 submitted to an open standards body.
- The project is released under Apache 2.0, built by SMJAI and contributors, and is actively recruiting adapter builders and SDK contributors for Go, Rust, and Java.
Why it matters: Power users juggling Claude, ChatGPT, Cursor, and custom agents gain a self-hosted, vendor-neutral way to carry preferences and context across every tool — replacing the current reality of repeating themselves each time they switch. For AI vendors, OMP introduces competitive pressure to ship compatible adapters or risk losing users who treat memory portability as a baseline expectation rather than a premium feature.
Ask SkimNews
