Companies that want Claude’s agents working inside their systems can now run the whole Claude Managed Agents service themselves, keeping agent records and credentials in-house.
Until now, customers could host only the sandbox where an agent runs its tools, while Anthropic handled the credentials and the record of what each agent did. Now an agent that triages alerts or fixes broken builds unattended can leave its full history on the company’s own servers. Anthropic announced the option in the 12 November entry of its Claude Platform release notes, and Help Net Security was first to report it.
You run the whole Managed Agents service, not only self-hosted sandboxes, on your own infrastructure, and you store agents, sessions, and session history there.
Anthropic, Claude Platform release notes
Help Net Security says the customer runs the Managed Agents server, its PostgreSQL database, and the workers where agents use shell and file tools. Session history lives in that database. The Next Web adds that agents, session histories, and credentials stay on the company’s systems, and every tool call can be logged.
Anthropic’s developer documentation says a persistent deployment requires PostgreSQL 16 or later; without a database, a restart erases the data. Developer write-ups describe Docker Compose for development and Helm charts for production.
The model still lives elsewhere
Self-hosting changes who runs the server, not where Claude runs. Each time an agent asks Claude something, the request still goes out to a model provider, and the answer comes back. The Next Web is blunt about it: Claude itself doesn’t run inside the customer’s network.
Customers do pick that provider. The server sends model requests to Anthropic by default, but the docs also list Amazon Bedrock, Claude Models on Google Cloud, Microsoft Foundry, or the company’s own model gateway.
| Part | Who runs it |
|---|---|
| Managed Agents server | The customer |
| PostgreSQL database with agents, sessions, and history | The customer |
| Workers running shell and file tools | The customer |
| Credentials and tool-call logs | The customer |
| The Claude model | Anthropic, Amazon Bedrock, Google Cloud, Microsoft Foundry, or the customer’s own gateway |
Some organisations already have a zero data retention (ZDR) agreement with Anthropic. Help Net Security says those customers’ model requests are covered the same way as any other Messages API call.
Running the service means taking on the jobs Anthropic used to do. The docs say customers run and secure a server, its database, and their sandboxes, and install each release themselves. Unite.ai adds hosting memory stores and detecting unusual usage to that list.
Nothing moves over from the hosted version either. Companies already using Anthropic’s cloud will have to recreate their agents, skills, vault credentials, and memory stores on the new server, according to Unite.ai. Some features are also missing, with The Next Web naming webhooks and scheduled runs.
Customers request access through their Anthropic account team. Datastudios says the release archive comes from that team, not a public download page, and Help Net Security describes the option as a private beta. Anthropic hasn’t published a price yet. Companies pay for their own infrastructure plus model usage with whichever provider they choose.
Why companies want it
The Next Web frames the move as giving cautious companies more control over where agent data lives, and points to Europe. There, European Central Bank president Christine Lagarde has warned about relying on foreign AI, and Mistral has pitched sovereign AI.
Anthropic launched Claude Managed Agents as a hosted service in April to take the infrastructure work out of running agents at scale. In September, OpenAI took a different route, letting companies run its agents inside Amazon’s cloud.
The option could suit Gulf companies with data-residency rules, but it needs checking. None of the sources says which regions it supports. Model calls still leave the network, so a buyer would need to ask the account team where its chosen provider processes those requests. Microsoft and Inception42 have already pitched UAE sovereign AI agents inside Microsoft 365 Copilot, and Microsoft Foundry is one of the providers on Anthropic’s list. A security team weighing agents can now keep the database and logs on its own servers, even if the model’s answers still arrive from outside.
