What Is Claude Managed Agents

What Is Claude Managed Agents

Most AI chatbots answer one message and stop. Claude Managed Agents is Anthropic’s service for the opposite kind of work — agents that keep going for hours, working through a bug backlog, fixing broken builds, testing releases and triaging alerts. And on November 12, 2026, Anthropic opened a private beta that lets companies run that entire service inside their own walls.

The announcement matters because it answers the objection that has kept autonomous agents out of banks, hospitals and law firms — “our data cannot leave the building.” With self-hosted Managed Agents, the agent service moves home. Only the model stays outside.

What Is Claude Managed Agents?

Claude Managed Agents is Anthropic’s managed platform for building and running long-running AI agents. Instead of every team hand-building its own agent loop, sandboxing, credential handling and session storage, Anthropic provides those pieces as composable APIs — the harness that runs the agent, the sandboxes where its tools execute, the vault that holds its credentials, and the history of every session.

It launched as a fully hosted beta in April 2026, running entirely on Anthropic’s infrastructure. In May, Anthropic added self-hosted sandboxes, so the workspace where an agent runs code could sit on the customer’s own machines. The November 12 private beta completes the move — now the whole service can be self-hosted.

What Actually Moves Into Your Own Cloud?

A self-hosted deployment has three parts, all operated by the customer. The first is the Managed Agents server, which serves the same API as the hosted service, runs the agent loop and stores agents, sessions and history. Notably, the server itself never runs an agent’s shell or file tools.

The second part is a database — PostgreSQL 16 or later — that holds the session history. Without it, everything lives in memory and a restart wipes it. The third part is the workers, which run the agents’ shell and file tools inside the customer’s own sandboxes.

Because the API is unchanged, the Claude SDK and the ant command-line tool connect to a self-hosted server exactly as they connect to Anthropic’s. Anthropic also says it has no remote access to a self-hosted server or its logs.

IT team monitoring AI agents on wall screens in an operations office

The Catch — The Model Does Not Move

Self-hosting moves the control plane, not the model. Every model call still sends the conversation so far — prompts, tool inputs and tool results — to an inference provider the customer chooses. The default is Anthropic’s Claude API, but teams can route calls through Amazon Bedrock, Claude models on Google Cloud, Microsoft Foundry, or a gateway they run themselves.

That distinction is the first thing a regulated business has to settle. Session history, credentials and files stay home, which is a genuine control gain and the only way to run Managed Agents under zero data retention. But what the model provider sees on each call is governed by that provider agreement, not by where the server sits.

The Trade-Offs of Self-Hosting

Self-hosting is a trade, not a free upgrade. Several hosted features are not yet available on the self-hosted server, including scheduled deployments, cloud environments, the Files API and webhooks, and web search and fetch stay off until explicitly enabled. Migration is also a fresh start — agents, skills, credentials and session history from the hosted service are not copied over and must be recreated.

There is an operational commitment too. Under the licence terms reported from the launch, customers must install Anthropic’s safety updates within one business day and roll back a flagged version within eight hours of a written safety notice. In return, the server software itself is royalty-free — model usage is billed under the customer’s separate Anthropic agreement.

It is the same bargain intelligent automation always offers at enterprise scale — more control, in exchange for more responsibility. And Anthropic is not alone in making it. Mistral moved its Studio platform to customer VPC and on-premises deployments on November 6, a sign that “agents inside your perimeter” is becoming an industry pattern, alongside model releases like Gemini 4 Hydrogen that make the agents themselves cheaper to run.

Bank security operations room running AI agents behind glass

Who Is It For, and How Do You Get It?

Anthropic describes the strongest fit as a regulated business with long-running agent work and a real perimeter to protect — a bank triaging alerts, a healthcare provider working through a bug backlog, or a law firm testing software releases. For those organisations, keeping credentials and session history inside their own network is often the difference between “not approved” and a pilot.

Access is limited for now. The self-hosted option is a private beta, available only to eligible organisations through their Anthropic account team. There is no public pricing and no general-availability date yet. For everyone else, the hosted Managed Agents service on the Claude Platform remains the way in — and understanding how AI is created is a good reminder of why the model itself still lives in the provider’s data centres, even when everything around it moves home.

Frequently Asked Questions

What is Claude Managed Agents?

Claude Managed Agents is Anthropic’s platform for building and running long-running AI agents — agents that work for hours on tasks like fixing bug backlogs, testing releases and triaging alerts. It provides the agent harness, sandboxes, credential handling and session history as managed APIs.

What changed with the self-hosted private beta on November 12, 2026?

Companies can now run the entire Managed Agents service — the server, agent loop, sandboxes, credentials and session history — on their own cloud or on-premises infrastructure, instead of on Anthropic’s. Before this, only the sandboxes could be self-hosted.

Does the Claude model itself run on the customer’s servers?

No. The service moves in-house, but model calls still go to a provider the customer chooses — Anthropic by default, or Amazon Bedrock, Google Cloud, Microsoft Foundry or the customer’s own gateway. Each call sends the conversation so far to that provider.

What does a company need to run self-hosted Claude Managed Agents?

Three things — the Managed Agents server, a PostgreSQL 16 or later database for session history, and workers that run the agents’ shell and file tools in local sandboxes. The customer also takes on patching duties, including installing safety updates within one business day.

What are the downsides of self-hosting Managed Agents?

Some hosted features are missing for now, including scheduled deployments, cloud environments, the Files API and webhooks. Nothing migrates from the hosted service, so agents, skills and credentials must be rebuilt, and the customer owns the server, database and firewall upkeep.

How can a company join the self-hosted beta?

Access is through an Anthropic account team, for eligible organisations only. Anthropic has published no public pricing and no date for general availability, so interested companies need to contact Anthropic directly.

Leave a Reply

Your email address will not be published. Required fields are marked *