Claude Agents Now Remember Across Sessions

Anthropic has launched built-in memory for Claude Managed Agents in public beta. The feature lets agents learn from every session using a filesystem-based memory layer, giving developers full control to export, manage, and audit what agents retain via API. Early adopters are already seeing results: Rakuten cut first-pass errors by 97%, Wisedocs sped up document verification by 30%, and Netflix agents now carry context across sessions without manual prompt updates. Memory supports shared stores, scoped permissions, and concurrent multi-agent access for enterprise deployments.