Anthropic Opens Self-Hosted Claude Managed Agents in Private Beta

Anthropic has moved Claude Managed Agents into a self-hosted private beta, letting enterprises run the agent harness inside their own cloud or on-premise environment instead of Anthropic's. Help Net Security reports the option opened on November 12, with access routed through Anthropic account teams.

In the hosted version Anthropic holds the agent infrastructure. In the self-hosted edition the customer runs it. "In the self-hosted setup, the customer operates three things: the Managed Agents server, its PostgreSQL database, and the workers where agents run their shell and file tools," the article explains. Session history lives in that database. The Claude model itself stays external, so prompts and completions still traverse the network, but model requests can be pointed at Anthropic directly, at Amazon Bedrock, at Google's Agent Platform, at Microsoft Foundry, or at an internal gateway.

Credentials are handled outside the sandbox where agent code runs. For MCP tools and supported API calls, a proxy sitting outside the sandbox attaches the credential at the moment of the request, and self-hosted customers can keep those secrets in their own secrets manager. Agents are restricted to allowlisted hosts, and the harness that calls Claude is kept separate from the sandbox that runs tool code.

Every step an agent takes is logged. "The log exports over OpenTelemetry, so security teams can pipe it into the monitoring tools they already run," the article notes. Customers pick up the operational load in return: patching, firewall rules and configuration all become their responsibility. The piece does not disclose pricing and no beta customers are named.

It lands the same day Google opened its Agent Gateway preview for EKS, AKS and OpenShift, part of a run of enterprise-agent infrastructure moves we have been tracking this week.