# AtomicMemory

A portable semantic memory layer for AI agents that gives them durable context across sessions, available as a managed hosted service, a self-hosted open-source core, a TypeScript SDK or an MCP server.

AtomicMemory stores and retrieves an agent's memory without locking an application to one model, framework or deployment: a developer can start with the managed hosted cloud, run the open-source core locally, or integrate through the TypeScript SDK and MCP server. Its CLI installs and initializes hosted cloud in one guided command, storing a per-installation credential so other machines are unaffected, or can be pointed at a local, self-hosted setup instead.

On its own published benchmarks, covering several memory-retrieval test sets, the project reports leading or parity performance against cost compared to published competitors. Basic usage stores a preference and retrieves it with simple ingest and search commands, and an integrate command connects the active memory profile to an agent host.

Website: https://docs.atomicstrata.ai