Anthropic announced a private beta of self-hosted Claude Managed Agents on November 12. The option is aimed at teams that want to run long-lived coding and operations agents against internal systems while keeping the agent service and its working records inside infrastructure they operate.
With the self-hosted setup, a customer operates the Managed Agents server, database and workers. The harness, sandboxes, credentials and session history can run in the customer’s cloud or on-premises environment. Developers use the same core APIs as Anthropic’s hosted service, while model requests can be routed to Anthropic, Amazon Bedrock, Google’s Agent Platform, Microsoft Foundry or a customer gateway.
Anthropic says the design separates the agent harness from the sandbox where code runs, keeps credentials outside the sandbox, supports network allowlists and records each session step for export through OpenTelemetry. These controls give operators more places to set policy and inspect activity; they do not mean every model request runs locally, since customers choose where requests are sent.
Self-hosting adds deployment control for organizations that could not use a hosted agent service for sensitive repositories or systems. The product remains in private beta, and Anthropic’s security descriptions are design claims from the vendor rather than an independent audit of customer deployments.