How is self-hosted Claude Managed Agents different from self-hosted sandboxes?

Self-hosted sandboxes move only the tool workspaces to the customer, while Anthropic still runs the Managed Agents service. Self-hosted Claude Managed Agents moves the service too: the API, the agent loop and the storage for agents, sessions and session history all run on the customer's infrastructure, and the model can sit at Anthropic or another chosen provider.

Does self-hosted Claude Managed Agents still send data to Anthropic?

Only if Anthropic is the chosen model provider. For each model call, the self-hosted server sends the conversation so far, including tool inputs and results, to the configured inference provider, which is the Claude API by default. Companies can instead route calls to Amazon Bedrock, Claude on Google Cloud, Microsoft Foundry or their own model gateway.

Who is responsible for security when Claude Managed Agents is self-hosted?

The customer. With self-hosted Claude Managed Agents, the company runs and secures the server, its database and the sandboxes, and installs each release itself. Anthropic's docs warn that the agent's bash tool is a real shell, so workers should run in a disposable container or VM, as the provided Docker Compose stack and Helm chart do.

Does self-hosted Claude Managed Agents have every hosted feature?

Not all of them. Anthropic says some Claude Managed Agents features exist only on the hosted service and keeps a 'Differences from the hosted service' page that lists the exceptions. The self-hosted server keeps the same API, so the SDK, the ant CLI and the rest of the Managed Agents documentation still apply.

Anthropic · 2026-11-12 · major

Self-hosted Claude Managed Agents — run the whole agent service in-house

In short

Anthropic now lets companies run the full Claude Managed Agents service on their own cloud or on-premise servers, in private beta. Agents, credentials and session history stay inside; model calls can go to Bedrock, Google, Foundry or a gateway.

Anthropic · 2026-11-12 · 2 sources checked

Claude article card for operating Claude Managed Agents inside your security perimeter

Claude Managed Agents can now run entirely inside a company's own network, with the model provider of its choice.

Quick facts

Maker	Anthropic
Availability	Private beta, via your Anthropic account team
Runs on	Your cloud account or on-premise data center
Components	cma-server, PostgreSQL 16+, workers, console
Model providers	Claude API, Bedrock, Google Cloud, Foundry, own gateway
Audit trail	Every session step, exported via OpenTelemetry

What is it?

Self-hosted Claude Managed Agents moves the whole Managed Agents service — the API, the agent loop and the storage for agents and sessions — onto a customer's own cloud account or data center. Until now, customers could self-host only the sandboxes where tools run, while Anthropic ran the service. It is in private beta and requested through an Anthropic account team.

How does it work?

A deployment has a server (`cma-server`) that serves the API, runs the agent loop and calls the model; a PostgreSQL 16+ database for agents, sessions and history; and workers that poll the server and run the agent's shell and file tools in sandboxes. The server keeps the same API as the hosted service, so the SDK and the `ant` CLI work against it. Model calls go to the Claude API by default, or to Amazon Bedrock, Claude on Google Cloud, Microsoft Foundry or a company's own gateway.

Why does it matter?

Teams that run agents unattended on their code and internal systems — fixing broken builds, triaging alerts, working through bug backlogs — often cannot let session history or credentials leave their environment. Anthropic says such teams previously had to build this platform themselves; now they run Anthropic's, but they also take on securing the server and installing each release.

Who is it for?

enterprise platform and security teams running agents on internal systems

Frequently asked questions

How is self-hosted Claude Managed Agents different from self-hosted sandboxes?: Self-hosted sandboxes move only the tool workspaces to the customer, while Anthropic still runs the Managed Agents service. Self-hosted Claude Managed Agents moves the service too: the API, the agent loop and the storage for agents, sessions and session history all run on the customer's infrastructure, and the model can sit at Anthropic or another chosen provider.
Does self-hosted Claude Managed Agents still send data to Anthropic?: Only if Anthropic is the chosen model provider. For each model call, the self-hosted server sends the conversation so far, including tool inputs and results, to the configured inference provider, which is the Claude API by default. Companies can instead route calls to Amazon Bedrock, Claude on Google Cloud, Microsoft Foundry or their own model gateway.
Who is responsible for security when Claude Managed Agents is self-hosted?: The customer. With self-hosted Claude Managed Agents, the company runs and secures the server, its database and the sandboxes, and installs each release itself. Anthropic's docs warn that the agent's bash tool is a real shell, so workers should run in a disposable container or VM, as the provided Docker Compose stack and Helm chart do.
Does self-hosted Claude Managed Agents have every hosted feature?: Not all of them. Anthropic says some Claude Managed Agents features exist only on the hosted service and keeps a 'Differences from the hosted service' page that lists the exceptions. The self-hosted server keeps the same API, so the SDK, the ant CLI and the rest of the Managed Agents documentation still apply.

Try it

Contact your Anthropic account team, then follow platform.claude.com/docs/en/managed-agents/self-hosted