GemNews

All stories · Hot News

Anthropic lets enterprises self-host Claude Managed Agents

Anthropic is giving businesses the option to run Claude Managed Agents on their own infrastructure instead of relying entirely on Anthropic to operate the service.

By Janie Nguyen · · 5 min read

Share: WhatsApp · Telegram · X · Facebook · LinkedIn · Email

Announced on November 12, 2026, the new self-hosted version is now available in private beta. Businesses can run the entire agent platform in their own cloud or on-premises environment, including the system that manages agents, the tools they use, their credentials, and their session history.

The key change is simple: Anthropic still provides the technology, but businesses can now operate the system themselves.

Anthropic's announcement

What has changed?

Previously, businesses had two main options.

They could use Anthropic's hosted service, where Anthropic managed the infrastructure behind their agents. Or they could run agent sandboxes on their own infrastructure, while Anthropic continued to manage the rest of the service.

Now, businesses can bring the entire platform into their own environment.

Deployment optionWho operates the agent platform?
Hosted Managed AgentsAnthropic
Self-hosted sandboxesAnthropic operates the platform; the business runs the agent's execution environment
Fully self-hosted Managed AgentsThe business

In the fully self-hosted setup, businesses operate the server, database, workers, sandboxes, credentials, and session history.

How to choose an agent environment

Does this mean businesses run Claude locally?

No. Running the agent platform and running the AI model are two different things.

Think of the agent platform as the workplace where an AI agent receives tasks, uses tools, and keeps track of its work. The model is the intelligence that helps the agent decide what to do.

With self-hosted Managed Agents, a business can run the workplace on its own infrastructure while still using a model from Anthropic or another provider.

Businesses can send model requests to:

  • Anthropic's Claude API
  • Amazon Bedrock
  • Google Cloud
  • Microsoft Foundry
  • Their own model gateway

The business chooses where those requests go. If it uses Claude API, prompts and tool results are still sent to Anthropic for model processing. If it uses another provider, the requests go there instead.

Data and access in self-hosted Managed Agents

Why would a business want to do this?

The main reason is control.

For companies whose agents work with internal code, business systems, or sensitive information, keeping the agent platform inside their own infrastructure can make it easier to apply their existing security and data policies.

With self-hosting, businesses can:

  • Control where session history is stored: The database and agent records stay in their own environment.
  • Manage access themselves: They decide which internal systems and services agents can reach.
  • Set their own security rules: They control network access, credentials, and the agent execution environment.

Anthropic says it does not have remote access to a customer's self-hosted server or its logs.

However, this does not automatically mean all data stays inside the business. Model requests still go to the inference provider selected by the customer.

Self-hosted data and access

More control also means more work

Self-hosting is not a free upgrade. Businesses must operate and maintain the infrastructure themselves, including the server, database, workers, security controls, and monitoring.

The self-hosted version also does not include every feature available in Anthropic's hosted service. Some hosted features, such as the Files API, scheduled deployments, and Dreams, are not available in the same way. Businesses may need to build their own alternatives.

Feature differences between hosted and self-hosted

What this means for AI agents

Anthropic's announcement reflects a shift in how businesses can deploy AI agents.

Until now, companies that wanted to use a managed agent platform generally had to let the provider operate much of the infrastructure, or build and maintain their own system.

Self-hosted Managed Agents offers another option: use Anthropic's agent technology without having to run the agent platform on Anthropic's infrastructure.

The model can still come from Anthropic or another provider. The business simply gets more control over the system that puts that model to work.

Ask AI next

  1. How can a business request access to the private beta of self-hosted Claude Managed Agents?
  2. Is hosted or self-hosted Claude Managed Agents the better fit for a small ecommerce team without DevOps staff?
  3. How is Claude Managed Agents priced, and does self-hosting change what a business pays Anthropic?
  4. Which cloud services and monitoring tools do teams use to run a self-hosted Claude Managed Agents setup?
  5. Where does Anthropic post updates on which hosted features, such as the Files API, reach self-hosted Managed Agents?

Discuss on GemCommunity