Anthropic has launched built-in memory for Claude Managed Agents in public beta. Agents can now persist and share knowledge across sessions using a filesystem-based memory layer, eliminating the need for custom memory infrastructure. Memory stores support scoped permissions, audit logs, versioning, and programmatic API access. Multiple agents can read/write shared stores concurrently. Early adopters include Netflix, Rakuten (97% reduction in first-pass errors), Wisedocs (30% faster document verification), and Ando. Memories are stored as files, giving developers full export and management control via the Claude Console or CLI.

3m read timeFrom claude.com
Post cover image
63 Impressions