2 min read
Add as a preferred source on Google

Anthropic Opens Private Beta for Self-Hosted Claude Managed Agents

Organizations can run the agent harness, sandboxes, credential handling and session history in their own environments, with network restrictions and OpenTelemetry logs.

A sealed compute cabinet sits beside heavy cable conduits / TokenPost.ai
A sealed compute cabinet sits beside heavy cable conduits / TokenPost.ai

Anthropic has opened a private beta for self-hosted Claude Managed Agents, letting organizations run the service’s agent infrastructure in their own environments and set controls for data, credentials and network access.

The self-hosted setup includes the agent harness, sandboxes, credential handling and session history, as well as the server, database and workers that run shell and file tools. The harness manages an agent’s work, while sandboxes provide the environment where it uses tools. Organizations can operate these components within their own cloud or on-premises infrastructure.

Teams can send model requests to Anthropic, Amazon Bedrock, Google Agent Platform or an internal gateway. The service uses the same core application programming interfaces as the hosted version, allowing developers to keep working with the same basic interfaces as they move the service into their own environment.

The security design separates the agent harness from its sandbox. Session history and vault secrets stay outside the sandbox, and organizations can keep credentials in their own secrets managers. They can also restrict network access to approved hosts, with requests to other hosts blocked.

Organizations can export records of agent activity through OpenTelemetry, a system for collecting and sharing operational data. For organizations with zero data retention agreements with Anthropic, the same policies can apply to requests sent to Claude.

Anthropic lists fixing broken builds, testing releases, triaging alerts and managing bug backlogs as potential uses. These examples involve agents working with software development and operational tasks, including shell and file tools.

The private beta follows Anthropic’s public beta launch of Claude Managed Agents, which offered hosted infrastructure for autonomous agent workflows and let developers run sandboxes on their own infrastructure. The new beta extends self-hosting to more of the service, including its server, database, workers and session history. Organizations can request access through their Anthropic account team.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…