Anthropic lets companies self-host its Claude Managed Agents service
A private beta lets customers run the agent server, database and tool workers in their own cloud or data center. Model calls still go to an outside provider.
Key takeaways
- Anthropic opened a private beta of self-hosted Claude Managed Agents on Nov. 12; access goes through Anthropic account teams.
- Customers host the main server, a PostgreSQL database and the tool workers in their own cloud or data center.
- Model calls still go to an outside provider such as Anthropic, Amazon Bedrock, Google or Microsoft Foundry.
- Anthropic's documentation says it has no remote access to the self-hosted server or its log.
- Customers must patch and secure the server themselves, and the hosted version's webhooks and scheduled runs are not included.
Companies can now apply to run Claude Managed Agents, Anthropic's service for tool-using AI agents that remember earlier sessions, on infrastructure they control. The self-hosted option works in a customer's own cloud account or on-premise, Help Net Security reports. The option opened as a private beta on Nov. 12, and access is granted on request via Anthropic account teams.
What changes for customers?
Until now, customers could host the sandbox, the isolated environment where an agent's code runs, themselves, while Anthropic ran the rest, including credential handling and the record of each agent's actions, according to Help Net Security. With self-hosting, three parts move to the customer: the main server, a PostgreSQL database behind it, and worker machines that execute the agents' shell and file tools. Session history is stored in that database, and developers use the same core APIs as the hosted service.
Anthropic's documentation says the company "has no remote access to the server or its log." It also says Anthropic issues and holds none of the credentials the server accepts. Under the hosted service, Anthropic stores session transcripts until a customer deletes them; with the self-hosted version, the customer stores them.
Does Claude run inside the customer's network?
No. Model requests still leave the customer's systems. Help Net Security reports that the customer picks where they go, including Anthropic itself, Amazon Bedrock, Microsoft Foundry, Google's Agent Platform or an in-house gateway. Either way, Claude does not run on the customer's own systems, so prompts leave and replies return. Anthropic's documentation says every call to the model includes the whole conversation up to that point, tool inputs and results among it, and that the server only connects to Anthropic when the Claude API is the chosen provider.
The documentation also notes that when calls go to the Claude API, they carry ID headers that let Anthropic attribute usage to a session, agent or environment, and that no setting turns them off. Anthropic says the IDs contain no names or content, though they record when each was created; the server's requests to the Claude API also carry a User-Agent header that Anthropic can use to count self-hosted usage.
What are the trade-offs?
Customers take on the work Anthropic used to do. Help Net Security notes that the server, the database and the firewall rules become the customer's to run, patch and configure correctly. The Next Web reports that agents and saved data cannot be transferred from the hosted service and must be set up again, and that webhooks and scheduled runs are among the features missing.
Help Net Security also reports that Anthropic's protection that keeps credentials out of the sandbox covers MCP tools and "supported API calls," so other services an agent reaches should be checked before the agent goes live. Each step an agent takes is logged and can be exported over OpenTelemetry to existing monitoring tools.
Why does it matter?
The Next Web says the option gives cautious companies more say over where agent data is kept, and links this to European concerns about relying on foreign AI, citing warnings from European Central Bank president Christine Lagarde. It notes that rivals such as Mistral have long promoted sovereign AI, and that OpenAI took a different route in September by letting companies run its agents inside Amazon's cloud.
Disclosure: NextFish's AI drafting desk uses Claude, a model made by Anthropic.
Questions readers are asking
What is self-hosted Claude Managed Agents?
A version of Anthropic's agent service that customers run on their own infrastructure, including the server, its database and the workers that run agents' tools.
Does the self-hosted version keep all data in-house?
Agents, sessions and credentials stay on the customer's systems, but each model call still sends the conversation to the chosen model provider.
Who can use it?
It is in private beta, and access is granted on request via Anthropic account teams, according to Help Net Security.
What features are missing?
The Next Web reports that webhooks and scheduled runs are not included and that agents and data cannot be moved over from the hosted service.
Sources
- #anthropic
- #claude
- #ai-agents
- #self-hosting
- #data-sovereignty
Get the day's top AI & tech stories on WhatsApp: Follow the NextFish channel
Get stories like this every morning.
Related
Agents3 min
OpenAI launches ChatGPT Wallet so its dots agents can buy for users
The payment feature starts with US ChatGPT Pro subscribers. Users approve each purchase with a biometric check, and agents get a payment token, not the real card number.
Agents3 min
Google opens Antigravity's /teamwork agents, claims math and code results
Google Research says its multi-agent system helped prove a 2013 conjecture and port large codebases to Rust, and offers 10,000 academics free access. Results are Google's own.
AnalysisAgents3 min
How big shopping and travel sites now write rules for AI agents
eBay, Target, Booking, Delta and Shopify have all added terms about AI agents. NextFish compared their terms with archived copies to see what changed and when.
Agents3 min
OpenAI's ChatGPT Wallet lets its dots agents buy things for users
Built with American Express, Mastercard and Visa, it starts with some US ChatGPT Pro users. Each purchase needs user approval, and OpenAI says agents never see card details.