Anthropic releases Memory API GA for Claude: structured persistent storage for agents across sessions

Picture hiring a highly capable assistant who, every single morning, walks through the door with no memory of anything that happened before. You have to re-explain who you are, what you are building, what decisions you made together yesterday. Until now, that was essentially how almost every AI agent built on an API worked: each conversation started completely fresh.

Anthropic's Memory API, now generally available, changes that. Developers can finally give their Claude-based agents structured, persistent memory that survives from one session to the next.

The API offers three tiers of memory: project-level (information shared across all users of the same application), user-level (preferences and context specific to each individual), and semantic search over stored facts, allowing the agent to surface relevant information even without knowing exactly what to look for.

What this means in practice: a customer support agent will remember each customer's full history. A coding assistant will remember your project conventions, your preferred libraries, and bugs you already resolved. An enterprise assistant will know who you are, your role, and the decisions made over recent weeks.

This is not just a technical convenience — it changes the mental model for designing agentic applications. AI APIs are no longer stateless tools to query one at a time; they become services with shared, continuous state. For anyone building AI applications, this is one of the most significant architectural shifts in recent years.