Anthropic Makes Claude Managed Agents Self-Hosted

This is a meaningful enterprise deployment upgrade, but “self-hosted” describes the control plane and execution environment—not private model inference. Teams gain stronger data residency and operational control while keeping Anthropic or another provider in the inference loop.

• –Keeps agent code, credentials, sandboxes, and session history inside the company’s security perimeter.
• –Supports provider choice and custom gateways, making the platform more compatible with existing model-routing strategies.
• –Separates the agent harness from tool execution, with network allowlists, secret-manager support, and OpenTelemetry tracing.
• –Developers should audit exactly what leaves the network: every model call includes the conversation, tool inputs, and tool results.
• –The private-beta implementation still lacks some hosted features, including scheduled deployments, webhooks, and several prebuilt skills. [Documentation](https://platform.claude.com/docs/en/managed-agents/self-hosted-data-and-access)