Anthropic now lets companies run its Claude Managed Agents service on their own servers, the company said in its release notes.[1] The self-hosted version is available now in private beta, and customers must request access through their Anthropic account team, according to the documentation.[2][3]
Until now, Anthropic hosted the orchestration and storage layers for Managed Agents, while customers could self-host only the sandboxes where tools ran; now they can run the entire service, including session histories and credentials, on their own infrastructure.[1][2] That means companies handling sensitive workflows can keep agent data under their control, though each model call still goes to an outside provider such as Anthropic, AWS, Google Cloud, or Microsoft Foundry.[1]
The move answers demand from cautious enterprises that want to run agents with access to internal systems without sending session data to the cloud provider.[1][2] It also aligns with sovereign AI concerns, especially in Europe, where reliance on foreign AI has drawn official scrutiny.[1]
Customers take on the work Anthropic used to do, such as securing and updating the server, and nothing moves over from Anthropic’s hosted service, so agents and saved data must be rebuilt.[1] Some features, including webhooks and scheduled runs, are missing in the self-hosted version, and each deployment must be isolated by trust level, meaning some organizations will need multiple servers.[1][2]