Anthropic opened a private beta on November 12 for self-hosted Claude Managed Agents, letting companies run the entire agent service — harness, sandboxes, credentials and session history — inside their own cloud or on-premises environment. Why it matters: agents that fix code and touch internal systems unattended are exactly the ones enterprises were least willing to run on someone else’s servers.
November 12, 2026
What happened
Anthropic announced that Self-hosted Claude Managed Agents is now available in private beta. Managed Agents is Anthropic’s suite of composable APIs for building and deploying autonomous agents, providing the agent harness, sandboxes, credential handling and session history needed to get long-running agents into production.
Until now, Managed Agents ran as a hosted service on the Claude Platform, with an option to self-host just the sandbox. With this release, every part of the service can run in the customer’s own environment.
What the customer runs now
In the self-hosted setup, the customer operates the Managed Agents server, its database, and the workers where agents run their shell and file tools. Session history is stored in the customer’s own database, and credentials can be kept in the customer’s own secrets manager.
Model requests go where the customer chooses: to Anthropic directly, through a cloud provider — Amazon Bedrock, Google’s Agent Platform or Microsoft Foundry — or through the customer’s own gateway. Developers still build agents with the same core APIs as the hosted service. If an organization has a zero data retention agreement with Anthropic, Anthropic says it applies to requests sent to Claude the same way it does for any other Messages API call.
Anthropic points to the kind of work driving the change: agents fixing broken builds, testing releases, triaging alerts and working through bug backlogs, running unattended on a company’s code with access to internal systems.
How the security model works
Anthropic says Managed Agents was designed around four principles, which apply in hosted and self-hosted setups alike. The harness (the “brain” that calls Claude) and the sandbox (the “hands” where agent code runs) are kept separate, so session history and vault secrets stay outside the sandbox. Credentials stay out of the sandbox too — a proxy outside it adds credentials to MCP and supported API calls. Network access can be limited to an allowlist of hosts, with other requests blocked. And every session leaves a full record of tool calls and results, exportable over OpenTelemetry into the monitoring tools a team already uses.
Why it matters
This is the enterprise trust problem in one product decision. The more useful an agent gets — the deeper it reaches into code, credentials and internal systems — the harder it is for a regulated or security-sensitive company to justify running that agent’s control plane on a vendor’s infrastructure. Self-hosting does not remove the model itself from the equation, since prompts still go out to Claude and completions come back, but it moves the operational layer — who holds the credentials, the session history and the audit log — inside the customer’s perimeter. For teams that were building their own agent platform just to get that control, this beta is a direct alternative. Access runs through Anthropic account teams, and no pricing or general-availability date was announced.
Frequently asked questions
Is self-hosted Claude Managed Agents generally available?
No. It launched in private beta on November 12, 2026. Companies request access through their Anthropic account team.
What runs on the customer’s infrastructure?
The Managed Agents server, its database, the workers that run agent shell and file tools, the sandboxes, credentials and session history.
Where do model requests go?
Where the customer points them: Anthropic directly, Amazon Bedrock, Google’s Agent Platform, Microsoft Foundry, or the customer’s own gateway. Claude itself still runs outside the customer’s network.
How is this different from self-hosted sandboxes?
Self-hosted sandboxes moved only tool execution in-house while Anthropic still ran the service. Self-hosted Managed Agents moves the whole service — control plane included — into the customer’s environment.
Sources: Anthropic (Claude), Help Net Security