Claude Code Sessions Can Now Run on Your Own Servers: Self-Hosted Environments in Beta!

Hey everyone, it's me, Shiichan! Today I found news about Claude Code getting a new place to live. Turns out you can now run sessions right on your own servers, so let's take a look!

Claude Blog claude.com

What was announced?

From Anthropic's Claude Blog: Claude Code now offers self-hosted environments in public beta. Sessions that used to run only on Anthropic-hosted servers can now execute on your own organization's infrastructure instead.

You can kick off sessions from web, mobile, desktop, or routine (scheduled) interfaces, and they run inside your organization's network alongside your internal services and security controls.

The story so far

Until now, Claude Code sessions ran on Anthropic-hosted servers. That meant if you wanted Claude to reach internal databases, services, or private package registries, you had to find some way to expose them externally.

Making sure compilers, SDKs, and custom CLIs were properly available every time was also a hassle, and the stricter an organization's security requirements were, the harder it was to fold Claude Code into internal workflows.

What changes

With self-hosted environments, here's what gets better:

• Sessions run inside your network perimeter, so you can reach internal databases, services, and registries directly without exposing them publicly
• You can pre-install compilers, SDKs, and custom CLIs, so every session starts with the tools it needs already in place
• Repository checkouts, build artifacts, and any files created or modified stay on your own infrastructure. Only conversation data, meaning prompts, responses, and tool results, is sent to Anthropic, and session transcripts are stored so you can pick up where you left off across surfaces

For platform teams who want to fold Claude Code into their existing development workflows while keeping their own security controls intact, this is exactly the option they've been waiting for.

Dive Deep

You can choose between two ways of running the compute (runners):

• Fixed: keep a constant number of runners running at all times, distributing session load across them
• On-demand: the orchestrator spins up runners as sessions queue up, then shuts them down once work is done, matching capacity to demand

Each session runs isolated in its own checkout, so there's no risk of data mixing between developers. There's also support for routing from multiple surfaces into the same shared environment.

This is available on Team and Enterprise plans only. It assumes a platform, developer experience, or productivity team owns setup and maintenance, including building and updating runner images and operating the orchestrator. It's not available for organizations using Zero Data Residency (ZDR).

George Jacob, Senior Engineering Manager at Faire, says self-hosted environments let them integrate Claude Code into existing development workflows while maintaining their security controls, with Claude generating PRs, fixing CI issues, and responding to workflow events using scalable compute that understands their codebase.

There's a related feature called Remote Control, which lets you continue a session running on your local machine from your phone or browser. Self-hosted environments are different: they run on shared infrastructure managed by a platform team, so any user can use them without machine-specific dependencies. Full documentation is available at code.claude.com/docs/en/self-hosted-environments.

Wrap-up

• Claude Code self-hosted environments are now in public beta, letting sessions run on your own infrastructure instead of Anthropic's servers
• You get direct access to internal databases and registries, plus pre-installed compilers and SDKs, while repos and build artifacts stay on your own infra
• Runners come in fixed or on-demand modes, and each session runs isolated
• It's Team/Enterprise-plan only, with setup and maintenance owned by your platform team; organizations using ZDR aren't supported
• A great new option for platform and DevOps teams at security-conscious organizations who want to weave Claude Code into their internal workflows