What happened
Claude’s agents can now live entirely behind your firewall, if you get into the beta. Anthropic opened a private beta of self-hosted Claude Managed Agents on 12 November 2026, letting companies run the whole agent service on their own cloud or on-premises infrastructure.
Until now, teams could move only the sandboxes onto their own machines. Now the entire service moves home: the agent harness, the server, the database, the workers that run shell and file tools, the credentials and the session history. You operate three pieces: a cma-server that serves the Managed Agents API, runs the agent loop and stores agents, sessions and history; a PostgreSQL 16 or later database that holds that history; and workers that run the agents’ shell and file tools inside your own sandboxes. The server itself never runs an agent’s shell or file tools.
Here is the catch, and it matters: the model itself does not move home. Every model call still goes to a provider you choose, Anthropic by default, or Amazon Bedrock, Claude models on Google Cloud, Microsoft Foundry, or your own gateway, set with the inference.provider setting. Anthropic says it has no remote access to your self-hosted server or its logs.
The beta is narrow. Access is only through an Anthropic account team, for eligible organisations. Anthropic has published no pricing, no list of covered countries and no date for general availability. Nothing migrates from the hosted service. Agents, skills, vault credentials, memory stores and session history must all be rebuilt. Some hosted features are missing in the beta, including scheduled deployments, dreams and webhooks, and web search and web fetch stay off until explicitly enabled. The development Docker Compose setup is explicitly not for production.
Also on 12 November, Anthropic started rolling out Claude Meetings in beta on Team and Enterprise. Claude joins a Google Meet call as a visible participant and builds docs live. That is a separate story; the how-to is in Trending, linked below.
Photo: NOIRLab/NSF/AURA/T. Slovinský, CC BY 4.0, via Wikimedia Commons
Why it matters for developers
This removes the blocker that kept Claude Code out of regulated teams: data residency. If your company could not let agent history, credentials and session logs live on Anthropic’s servers, you now have a path. The harness, the sandboxes and the history sit in your cloud, under your control.
But read the fine print before you celebrate. Model traffic still leaves your network to whichever provider you pick, so this is self-hosted orchestration, not an on-premises model. If your compliance team needs the model itself inside the firewall, this beta does not give you that, and the article says so plainly because that distinction is the whole story.
Nothing breaks today for everyone else. The hosted service keeps running exactly as before, and the beta is private. If you are not in it, your Monday looks like your Friday.
What changes for Claude Code / Cursor / Codex users
- Claude Code: if your team is in the beta, this is the biggest enterprise change of the year: long-running agents on internal code, with history and credentials inside your walls. If you are not in the beta, nothing changes, and you cannot sign up for it yourself; it goes through your Anthropic account team.
- Cursor: nothing changes directly. But watch this space: the enterprise self-hosting bar has just been set, and Cursor’s enterprise story will be judged against it.
- Codex: nothing changes directly. The same judgement applies. Self-hosted agent infrastructure is now a competitive feature, not a nice-to-have.
What to watch next
- Beta access: through your Anthropic account team only, for eligible organisations. There is no public signup page and no published pricing.
- General availability: no date has been published. Treat any date you hear as provisional until Anthropic writes it down.
- The missing features: scheduled deployments, dreams and webhooks are unavailable in the beta. If your workflow depends on them, the hosted service is still the right choice.