Anthropic Launches Private Beta for Self-Hosted Claude Managed Agents

Anthropic has announced the private beta launch of self-hosted Claude Managed Agents, a significant expansion of its Managed Agents service. This new offering allows development teams to operate the full suite of Managed Agents capabilities within their own cloud or on-premises infrastructure, addressing the growing demand for enhanced security and control over sensitive data and internal systems.

Key Features of Self-Hosted Claude Managed Agents

The self-hosted version of Claude Managed Agents enables organizations to run every component of the service within their own environment. This includes the agent harness, sandboxes, credential handling, and session history. Teams can now operate the Managed Agents server, its database, and the workers responsible for running shell and file tools. Model requests can be routed to Anthropic, a preferred cloud provider such as Amazon Bedrock or Google Agent Platform, or even an organization’s own gateway.

Developers retain access to the same core APIs used in the hosted Managed Agents service, ensuring continuity in building and deploying autonomous agents. Additionally, session history is stored locally, and organizations with zero data retention (ZDR) agreements with Anthropic can apply the same policies to requests sent to Claude.

Enhanced Security Architecture

Anthropic has designed the self-hosted Managed Agents with robust security principles to ensure safe operation of autonomous agents within enterprise environments:

• Separation of components: The harness (agent brain) and sandbox (agent hands) are kept distinct, ensuring that session history and vault secrets remain outside the sandbox.
• Credential management: Credentials are excluded from the sandbox and can be stored in an organization’s own secrets manager.
• Network access controls: Organizations can restrict agent network access to specific hosts, with any unauthorized requests being blocked.
• Comprehensive session logging: Every agent action is recorded and can be exported via OpenTelemetry for integration with existing monitoring tools.

Applications and Availability

Self-hosted Claude Managed Agents are particularly suited for tasks requiring deep integration with internal systems, such as fixing broken builds, testing releases, triaging alerts, and managing bug backlogs. By operating entirely within an organization’s infrastructure, the service supports compliance with strict data security and audit requirements.

The private beta is now open, and interested organizations can contact their Anthropic account team to request access or consult the documentation for further details.