Anthropic adds self-hosting option for Claude Managed Agents
Companies can keep agent histories and credentials on their infrastructure, but requests to Claude still go to an external model provider.
Anthropic has introduced a self-hosted version of Claude Managed Agents, allowing businesses to operate the service within a cloud account or data center they control. Customers need to contact their Anthropic account team for access, the company said in its release notes.
The service runs AI agents that can use tools and retain information across sessions. Previously, customers could host the tools themselves, while Anthropic operated the remaining components. The new option keeps agents, session records and credentials on customer infrastructure, with logging available for every tool call.
Self-hosting does not put the Claude model inside a customer's network. Conversations are still sent externally whenever an agent makes a model request. Customers can select a provider, with options including Anthropic, Amazon Bedrock, Google Cloud and Microsoft Foundry.
Businesses also become responsible for server security and updates, work previously handled by Anthropic. Existing agents and stored data do not transfer from the hosted service and must be recreated. The self-hosted version also lacks some features, including webhooks and scheduled runs.
