Claude Managed Agents have launched a public beta for their memory feature, allowing agents to learn from every session. This innovation utilizes a memory layer optimized for intelligence, ensuring a balance between performance and flexibility.
Memory is designed to integrate seamlessly with existing tools, mounting directly onto a filesystem. This enables Claude to leverage bash and code execution capabilities, enhancing its effectiveness in agentic tasks. The new filesystem-based memory allows for more comprehensive and organized memory storage, improving the agent's ability to discern what to remember for specific tasks.
This feature is particularly beneficial for enterprise environments, providing scoped permissions, audit logs, and programmatic control. Memory stores can be shared among multiple agents with varying access levels. For instance, an organization-wide store could be read-only, while individual user stores allow for both reading and writing.
Developers can manage memories as files through the API, maintaining complete control. Each memory change is tracked with detailed audit logs, making it easy to identify the source of each memory and revert to previous versions if necessary. Updates appear in the Claude Console as session events, allowing developers to monitor what agents have learned.
Teams utilizing this memory feature have reported significant improvements in their workflows. For example, agents are able to close feedback loops more effectively, expedite verification processes, and eliminate the need for custom retrieval infrastructures. This leads to a reduction in first-pass errors and overall operational costs.
Memory on Managed Agents is currently available in public beta on the Claude Platform. Developers can begin deploying agents with memory using the Claude Console or the new CLI. Additional documentation is available for those looking to explore the capabilities further.
Key Benefits of Memory in Managed Agents
- Continuous learning from every session.
- Improved performance metrics, including reduced errors and latency.
- Enhanced control over memory management and access.
What to Expect
As this feature evolves, users can anticipate further enhancements in memory capabilities, making it easier to implement continuous learning across various applications.
