Anthropic just launched memory for Claude Managed Agents in public beta.
Agents can now retain context across sessions, learning from past interactions, corrections, and discoveries without starting from scratch each time.
The memory is file-based. Claude saves memories to a filesystem it can read and write with bash and code execution, the same tools it already uses for agentic tasks. Memories can be exported, managed via API, and audited with full version history.
The enterprise angle is where it gets interesting. Memory stores can be shared across multiple agents with scoped permissions, an org-wide store can be read-only while per-user stores allow writes. Multiple agents can work concurrently against the same store without overwriting each other.
Early results from teams already using it: Rakuten cut first-pass agent errors by 97%. Wisedocs sped up document verification by 30%. Netflix is using it to carry insights across sessions instead of manually updating prompts.
Agents can now retain context across sessions, learning from past interactions, corrections, and discoveries without starting from scratch each time.
The memory is file-based. Claude saves memories to a filesystem it can read and write with bash and code execution, the same tools it already uses for agentic tasks. Memories can be exported, managed via API, and audited with full version history.
The enterprise angle is where it gets interesting. Memory stores can be shared across multiple agents with scoped permissions, an org-wide store can be read-only while per-user stores allow writes. Multiple agents can work concurrently against the same store without overwriting each other.
Early results from teams already using it: Rakuten cut first-pass agent errors by 97%. Wisedocs sped up document verification by 30%. Netflix is using it to carry insights across sessions instead of manually updating prompts.
31❤️1🤖1🤙1
