Anthropic adds built-in memory to Claude Managed Agents
Anthropic announced built-in memory for Claude Managed Agents on April 23, 2026, and says the feature is available in public beta on the Claude Platform.

Key takeaways · 4
- 01
The feature is in public beta on the Claude Platform.
- 02
Memory is mounted directly onto a filesystem, and developers can export and manage its files through the API.
- 03
Developers can control what agents retain and can roll back memory versions or redact content from memory history.
- 04
Rakuten reported 97% fewer first-pass errors, with cost 27% lower and latency 34% lower.
A filesystem-based memory layer
Anthropic says the memory layer mounts directly onto a filesystem and is optimized against its internal benchmarks for long-running agents.[1] Claude can use the same bash and code-execution capabilities that Anthropic says make it effective at agentic tasks.[1] The company says its latest models save more comprehensive, well-organized memories and are more discerning about what to remember for a given task.[1]
Memory across sessions and agents
Anthropic says agents using memory can learn from every session and share what they have learned with one another.[1] Netflix agents, for example, carry context across sessions, including information uncovered over multiple turns and human corrections.[1] Memory stores can also be shared among multiple agents with different access scopes.[1] Anthropic says an organization-wide store can be read-only while per-user stores allow reading and writing, giving developers options for separating shared context from user-specific updates.[1] Multiple agents can work concurrently against the same store without overwriting one another.[1]
Developer controls and visibility
Developers can control what agents retain, and memory files can be exported and managed through the API.[1] Anthropic says developers can identify the agent and session associated with a memory, roll back to an earlier version, and redact content from memory history.[1] Memory updates appear in the Claude Console as session events.[1] Developers can get started through the Claude Console or Anthropic’s new CLI.[1] These controls give teams ways to inspect and manage memory, while leaving them responsible for deciding what agents should retain and how stores should be scoped.[1]
Early customer examples
Anthropic says Rakuten uses memory in task-based, long-running agents to learn from sessions and avoid repeating past mistakes.[1] Rakuten says its agents delivered 97% fewer first-pass errors; cost was 27% lower and latency was 34% lower.[1] Wisedocs used cross-session memory to spot and remember recurring document issues, and document verification sped up 30%.[1] Ando is building a workplace messaging platform on Managed Agents to capture how each organization interacts.[1] Its founder, Sara Du, said memory lets the company stop building memory infrastructure and focus on its product.[1]
Persistent, shareable memory could reduce repeated setup and help agents incorporate corrections across sessions. Developers considering the beta should evaluate the access scopes and retention controls alongside the task benefits reported by Anthropic’s customers.
Why it matters
Bring this to your team — one session a day, built for your industry.
kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.
For teamsHow this developed
23 October 2026
Anthropic adds built-in memory to Claude Managed Agents