Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarTNW→ original

Anthropic Lets Companies Run Claude Managed Agents on Their Own Servers

Anthropic has let companies deploy Claude Managed Agents in their own cloud or data center. Agent session histories and access keys now stay on the customer's side, and every tool call can be written to a log. But requests to the model itself still go to an external provider: Anthropic, Amazon Bedrock, Google Cloud, or Microsoft Foundry. So there is no full isolation, but control over data storage has grown.

AI-processed from TNW; edited by Hamidun News
Anthropic Lets Companies Run Claude Managed Agents on Their Own Servers
Source: TNW. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTNW ↗
Editor-in-Chief
Key metric #1Access is by request only, through an Anthropic account manager
There is no migrationagents and stored data will have to be rebuilt from scratch
Key metric #3There are no webhooks or scheduled runs, and updates are the customer's responsibility

Corporate IT departments have long asked AI vendors the same question: where does an agent's data physically live? An agent remembers past sessions, accesses databases and services, and holds access keys. For a bank, a hospital, or a government agency, handing all of that to someone else's cloud is a serious risk. Anthropic is answering that request: on Thursday, the company launched a self-hosted version of Claude Managed Agents.

What moves to the customer's side

Managed Agents is Anthropic's service for running AI agents that can use tools and remember past sessions. Previously, customers could already run agents' tools on their own machines. Everything else, including stored history and credentials, was handled by Anthropic.

Now the entire service can be deployed in the customer's own cloud account or in its own data center. Agents, session histories, and credentials remain within the company's infrastructure, and every tool call can be written to a log. For auditing and compliance, this is a significant advantage: companies can show inspectors exactly what the agent did and with what permissions.

Access is not self-service. Customers have to request it from their Anthropic team.

Where the catch is

The main limitation: Claude itself does not run inside the customer's network. Each time an agent calls the model, the conversation goes to the model provider. That can be Anthropic itself, Amazon Bedrock, Google Cloud, or Microsoft Foundry. The customer chooses, but a fully closed loop is not possible.

In simple terms, the company takes over the agent's "memory and keys," but it still rents the "brain." If regulatory requirements prohibit sending dialogue content outside the perimeter, this option does not resolve the question entirely. If the main concern is storing histories, tokens, and keys, along with transparency of actions, it addresses that more strongly than the cloud version does.

The price of control

Self-hosting means that work Anthropic previously handled will fall to the customer: securing the server and installing updates. This requires a team with experience in security and operations.

There are also practical drawbacks. Nothing carries over automatically from the Anthropic cloud version, so agents and accumulated data must be rebuilt. In addition, the self-hosted version does not yet include some features, such as webhooks and scheduled runs. If your agents react to external events or run on a timer, you will need workarounds or will have to wait.

Market context

The move fits into the European agenda on digital sovereignty. ECB President Christine Lagarde has warned about dependence on foreign AI, and competitors such as Mistral have long sold the idea of control over one's own infrastructure. In September, OpenAI took a different route and also offered companies the ability to run its agents on their own premises, while Nvidia is building its own solutions for this market.

What to do in practice

Teams choosing a platform for agents should split their requirements into two lists. The first contains what must stay in-house: session histories, keys, and action logs. The second contains what can be handed to the model provider. If the first list fits within the new version's capabilities, a pilot makes sense. Assess in advance who will update and secure the server, and keep in mind that moving off the cloud version means rebuilding from scratch. If you need webhooks and schedules, it is currently safer to stay on the cloud service.

Story

Key Questions Answered

What moves to the customer's side?

Managed Agents is Anthropic's service for running AI agents that can use tools and remember past sessions. Previously, customers could already run agents' tools on their own machines. Everything else, including stored history and credentials, was handled by Anthropic. Now the entire service can be deployed in the customer's own cloud account or in its own data center. Agents, session histories, and credentials remain within the company's infrastructure, and every tool call can be written to a log.

Where the catch is?

The main limitation: Claude itself does not run inside the customer's network. Each time an agent calls the model, the conversation goes to the model provider. That can be Anthropic itself, Amazon Bedrock, Google Cloud, or Microsoft Foundry. The customer chooses, but a fully closed loop is not possible.

What are the key details on The price of control?

Self-hosting means that work Anthropic previously handled will fall to the customer: securing the server and installing updates. This requires a team with experience in security and operations. There are also practical drawbacks. Nothing carries over automatically from the Anthropic cloud version, so agents and accumulated data must be rebuilt. In addition, the self-hosted version does not yet include some features, such as webhooks and scheduled runs.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…