Claude Managed Agents Memory Goes Public Beta: Agents Can Now Remember Across Sessions

Claude Managed Agents Memory Goes Public Beta: Agents Can Now Remember Across Sessions

On April 23, 2026, Anthropic announced that the memory feature for Claude Managed Agents has entered public beta. This means Agents built through the Claude Platform no longer “live only in the present”—they can learn from each session, save experiences as files, and reuse them across individual or cross-session contexts.

How the Memory Layer Works

Anthropic’s design for the memory feature has several key points:

  • File-based storage: Memory is saved as files that users can intuitively view, edit, and manage
  • Cross-session persistence: Agents can access previously saved memory across different sessions
  • API exportable: Developers can export Agent memory files via API for auditing and analysis
  • Performance vs. flexibility balance: Anthropic describes the memory layer as “agent-optimized,” balancing retrieval performance with memory flexibility

Why This Matters

Before Claude Managed Agents, Anthropic’s Messages API was stateless—each call was an independent conversation with no persistent memory. This meant:

  • Building Agents requiring long-term context (e.g., project management assistants, codebase maintenance Agents) required developers to implement their own memory layer
  • Cross-session Agent behavior consistency was difficult to guarantee
  • Memory management became an additional engineering burden for building high-quality Agents

Now that Anthropic has built this capability into the Platform layer, it lowers the barrier to Agent development.

Comparison with Competitors

DimensionClaude Managed AgentsOpenAI Assistants APICustom Memory
Built-in Memory✅ Public Beta✅ Available❌ Self-built
Memory FormatFile SystemVector StorageCustom
Cross-SessionDepends
Auditability✅ Files ExportableLimitedDepends

Action Items

  • Teams already using Claude Platform: Try the memory feature and evaluate improvements to existing Agent workflows
  • Teams with custom memory layers: Compare the built-in solution with custom approaches on cost, performance, and maintainability
  • Teams not yet on Managed Agents: Evaluate migration feasibility from Messages API, with memory as a key differentiator

Primary Sources