Anthropic has opened a private beta of self-hosted Claude Managed Agents for organisations that need session data and agent management inside their own infrastructure. Customers host the service in their cloud or data centre, while model requests still go to a chosen provider. The announcement separates control over agent infrastructure from the location of model processing.
Anthropic has opened a private beta of self-hosted Claude Managed Agents, a service for building and running AI agents. Organisations can host agent management, session history and credentials in their own cloud or data centre. Model requests continue to a chosen provider. Access to the beta is arranged through Anthropic’s account team, with no general availability date given in the announcement.[1], [2]
The customer operates the service’s management components, application interface server, database and worker processes. Anthropic says developers use the same core application programming interface as the hosted service. They can send model requests directly to Anthropic, through Amazon Bedrock or services from Google and Microsoft, or through their own gateway. Existing zero data retention arrangements apply to those model requests.[1]
The security design separates the agent loop from the isolated environment where code runs. A proxy outside that environment can add credentials to requests, while organisations can retain secrets in their own managers. Host allowlists and firewall rules restrict the external services an agent can contact. Events and traces can be exported through OpenTelemetry, a monitoring framework, into customers’ existing monitoring systems.[1]
