# Anthropic opens self-hosted Claude Managed Agents to private beta

Published 2026-11-13 · AI.info Team · AI.info

> Organizations can run Claude Managed Agents, including its server, database and sandboxes, in their own cloud or on-premises. The private-beta service lets customers choose an inference provider while keeping session history in their own database.

Organizations can now host the Claude Managed Agents service and its sandboxes on their own infrastructure through a private beta. The option extends [Anthropic](https://claude.com/resources/articles/self-hosted-claude-managed-agents)'s earlier self-hosted sandbox offering to include the service that runs the agent loop and stores session history.

Anthropic previously offered a hosted service and an option to run only the sandboxes on infrastructure controlled by the customer. Developers can use the new option to build agents that work with company code and internal systems.

## What customers operate

The product is a way to run Managed Agents, a set of APIs for building and deploying autonomous agents. Organizations operate the server, its database and the workers that run agents' shell and file tools, either in their own cloud environment or on-premises.

The server handles the API and agent loop and stores agents, sessions and session history. Workers take tasks from the server and execute shell and file tools in sandboxes. Developers use the same core APIs as the hosted service.

The model runs at Anthropic or a provider selected by the customer. Requests can go directly to Anthropic, through a cloud provider such as [Amazon](https://ai.info/companies/amazon) Bedrock, [Google](https://ai.info/companies/google)'s Agent Platform or [Microsoft](https://ai.info/companies/microsoft) Foundry, or through the customer's own gateway.

For each model call, the server sends the conversation so far, including tool inputs and results, to the inference provider. An existing zero data retention arrangement with Anthropic applies to requests sent to Claude in the same way as other Messages API calls.

## Credentials and session records

The agent's code runs inside a sandbox, while the agent loop, session history and vault secrets remain outside it. Sandboxes start when a tool needs them, and session history is kept as an event log outside the sandbox.

For MCP tools and supported API calls, a proxy outside the sandbox adds credentials to requests. Self-hosted users can also store credentials in their own secrets manager. When hosting a sandbox themselves, organizations set network limits through their own firewall.

Organizations running self-hosted Managed Agents are responsible for securing the server, its database and sandboxes, and installing each release themselves.

Managed Agents records the steps taken during a session, including tool calls and their results. The record can be exported over OpenTelemetry to monitoring tools.

## Limits and access

Some features of the hosted service are not available on the self-hosted server, including scheduled deployments and webhooks. Web search and web fetch are off by default and require a server setting to enable them; their availability also depends on the inference provider.

Moving to a self-hosted server does not copy material from the hosted service, including session history. Agents, skills, vault credentials and memory stores must be created again on the server. To request access to the private beta, organizations can contact their Anthropic account team.

## Sources

1. Original source [Anthropic · Self hosted claude managed agents](https://claude.com/resources/articles/self-hosted-claude-managed-agents) claude.com

Other pages read

2. [Anthropic · page 2](https://platform.claude.com/docs/en/managed-agents/self-hosted) platform.claude.com
3. [Anthropic · page 3](https://platform.claude.com/docs/en/managed-agents/self-hosted/differences) platform.claude.com

## Companies in this story

- [Anthropic](https://ai.info/companies/anthropic)

## Related stories

- [Claude Meetings rolls out in beta for Team and Enterprise plans](https://ai.info/news/claude-meetings-beta-team-enterprise-google-meet)
- [Anthropic advises deleting "think hard" lines from Claude Opus 5.5 prompts](https://ai.info/news/anthropic-claude-opus-5-5-prompting-habits)
- [Anthropic resolves Claude errors after API and Code disruption](https://ai.info/news/anthropic-claude-api-code-errors-resolved)
- [Anthropic cuts vulnerability disclosure windows to 30 and 45 days](https://ai.info/news/anthropic-shortens-vulnerability-disclosure-timelines)
- [ElevenLabs previews enterprise agents and model fine-tuning](https://ai.info/news/elevenlabs-enterprise-agents-model-fine-tuning)

## Explore

[More articles](https://ai.info/news)

---

Canonical: https://ai.info/news/anthropic-self-hosted-claude-managed-agents-private-beta
