New featureReleased 12 November 2026 · 4 min read
Claude agents can run entirely inside your own perimeter.
Anthropic has launched self-hosted Claude Managed Agents in private beta. A company can now run the whole Managed Agents service in its own cloud or on-premise, including the part that calls Claude, the sandboxes, credentials and session history. Access is by request through your Anthropic account team.
At a glance
- Status
- Private beta
- Where it runs
- Your cloud or on-premise
- Session history
- Your own database
- Model requests
- Anthropic by default
- Access
- Via your Anthropic team
01 · What changed
What Anthropic shipped.
Managed Agents launched as a hosted service on the Claude Platform, where Anthropic runs the infrastructure. Anthropic then added self-hosted sandboxes for teams whose code had to stay in their own environment. Self-hosting now covers every part of the service: you operate the Managed Agents server, its database and the workers where agents run their shell and file tools. Anthropic says teams that wanted all of this in their own environment previously had to build the platform themselves.
- 1
The whole service, self-hosted
The part that calls Claude, the sandboxes, credentials and session history all run in your environment. You operate the server, its database and the workers.
- 2
Model requests go where you say
Requests go to Anthropic by default, or to Amazon Bedrock, Google's Agent Platform, Microsoft Foundry or your own gateway.
- 3
Same core APIs
Developers build agents with the same core APIs as the hosted Managed Agents service.
- 4
History in your database
Agents, sessions and session history are stored on your infrastructure. If you have zero data retention with Anthropic, it applies to requests sent to Claude as it does for any Messages API call.
- 5
Credentials kept out
Vaults keep credentials out of the sandbox, and a proxy outside it adds them to MCP and supported API calls. Self-hosting also lets you keep credentials in your own secrets manager.
- 6
Allowed hosts only
Each environment can be limited to a list of allowed hosts, and other requests are blocked. When you host the sandbox, you set these limits in your own firewall.
- 7
A record of every step
Every tool call and its result is recorded. The record exports over OpenTelemetry to the monitoring tools you already use.
02 · What it means
What it means for your seat.
For most firms this changes nothing about how people use Claude day to day. It matters to the teams building agents that run unattended against internal code and systems, and to the people who sign off on where those agents, their credentials and their records live. Because it is a private beta, the first step is to decide whether you need it and, if so, to request access.
Leadership
If agents have been held back because the agent service and its history had to sit outside your environment, that constraint now has an answer in beta. Decide whether any planned agent work, such as triaging alerts on internal systems, warrants running the whole service in-house, or whether hosted Managed Agents or self-hosted sandboxes are enough.
Risk and compliance
Session history, credentials and the record of every agent step can now stay on your own infrastructure, with the record exported over OpenTelemetry to your existing monitoring. Anthropic says an existing zero data retention arrangement applies to requests sent to Claude in the same way as any Messages API call. Confirm which model provider your engineers intend to route to, since that is where model requests go.
IT and engineering
Self-hosting means you operate the Managed Agents server, its database and the workers that run shell and file tools. Plan the allowed-hosts list for each environment (for a build-fixing agent, Anthropic's example is just your git host and package registry) and decide whether credentials live in vaults or your own secrets manager. Agents are built with the same core APIs as the hosted service.
Operations
Owning the service means owning its uptime, database and workers. Agree who runs it, who reviews the exported agent records and how agent activity sits alongside the rest of your system monitoring before anything runs unattended.
03 · This week
What to do this week.
0 of 3 done
Sources
Facts and dates are Anthropic’s own. What it means and what to do are our view, from the firms we work with. Drafted with Claude from these sources, checked against them figure by figure, and published 12 November 2026.
The short answers.
What is self-hosted Claude Managed Agents?
It is a way to run the whole Claude Managed Agents service, including the part that calls Claude, the sandboxes, credentials and session history, in your own cloud or on-premise. Anthropic has made it available in private beta.
How do I get access to self-hosted Claude Managed Agents?
It is in private beta. Anthropic says to contact your Anthropic account team to request access.
Where do model requests go when Managed Agents is self-hosted?
Model requests go to the provider you choose, which is Anthropic by default. Anthropic says you can also route them through Amazon Bedrock, Google's Agent Platform, Microsoft Foundry or your own gateway.
Does zero data retention apply to self-hosted Managed Agents?
Session history is stored in your own database. Anthropic says that if your organization has zero data retention in place with Anthropic, it applies to requests sent to Claude in the same way as any other Messages API call.
Read next
Newsletter
Keep up with every release.
One email when something ships or retires that matters to a company: what changed, what it means, the date and what to do. Unsubscribe any time.
Want this working inside your firm?
We run Claude workshops on your own documents and stay on as your teams adopt each release.