Anthropic Launches Native Memory for Claude Managed Agents to Enable Persistent Learning

ClaudeClaude

· Updated

Anthropic introduced a native memory layer for Claude Managed Agents in public beta, allowing autonomous systems to retain knowledge across multiple sessions. By storing memories as manageable files, the update removes the need for custom state-management infrastructure while giving developers full control over what an agent remembers.

Anthropic launched a native memory layer for Claude Managed Agents in public beta, enabling autonomous systems to learn from every interaction. This intelligence-optimized layer provides a built-in way for agents to carry context across sessions. It builds on the standardized production infrastructure introduced earlier this month.

This update extends the functionality of external session store adapters by moving the memory layer directly into the managed infrastructure. This shifts the burden of state management to the platform, providing a turnkey solution that balances real-time performance with the flexibility to recall specific user preferences or past outcomes.

You can now view, edit, and delete agent memories directly through the Anthropic console or via API. This level of control follows the company's research into autonomous negotiation agents that must retain strategy across sessions. The feature is currently available in public beta for managed agents users.

View the full update on claude.com
Claude
Claude
@claudeai
X

Memory on Claude Managed Agents is now in public beta. Your agents can now learn from every session, using an intelligence-optimized memory layer that balances performance with flexibility. https://t.co/P7GjOYPqCz

378retweets5.7klikes
View on X

Still wondering? A few quick answers below.

Memory on Claude Managed Agents is a native feature that allows AI agents to retain information and learn from every interaction across different sessions. Instead of starting with a blank slate each time, agents use an intelligence-optimized memory layer to maintain context, enabling more personalized and consistent behavior over long-term usage.

The system uses an intelligence-optimized memory layer designed to balance real-time performance with the flexibility to store complex details. It captures key information from agent sessions and persists it automatically. This allows autonomous agents to build a cumulative knowledge base that informs their reasoning and actions in future interactions without manual context injection.

Developers have full control over what agents retain because memories are stored as files. These files can be viewed, edited, or deleted directly through the Anthropic console or managed programmatically via the API. This transparency allows teams to export memory data for auditing, debugging, or fine-tuning the agent's long-term knowledge base.

Memory for Claude Managed Agents is currently in public beta. It is available to users who have access to the Claude Managed Agents platform, which provides the underlying infrastructure for hosting and scaling autonomous AI systems. Users can begin testing the persistent learning capabilities through the Anthropic developer console or the platform's API.

Native memory eliminates the need for developers to build custom infrastructure for state management or external databases to track user preferences. By integrating persistence directly into the managed agent layer, Anthropic simplifies the process of creating agents that improve over time while maintaining high performance and developer-led control over data retention.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update