Image: Anthropic / unite.ai Claude Code Now Runs on Your Own Servers
Anthropic's self-hosted environments for Claude Code entered public beta on August 6. Team and Enterprise customers can now run coding agent sessions on their own infrastructure, keeping source code and credentials off Anthropic's servers.
Anthropic put Claude Code’s self-hosted environments into public beta on August 6. Organizations on Team and Enterprise plans can now run their Claude Code sessions on their own machines or containers instead of Anthropic’s infrastructure.
The pitch is straightforward: if your team builds against internal services that aren’t reachable from the public internet, or if your security policy requires source code to stay on premises, you couldn’t use Claude Code’s cloud-based sessions before. Now you can.
How it works
You deploy processes called runners — long-lived daemons that wait for sessions to arrive and then spin up a Claude Code process for each one. Sessions come in through the claude.ai web interface, mobile apps, or the desktop client, but execute inside your network.
Two deployment modes are available. Fixed mode keeps a set number of runners running at all times; incoming sessions are distributed across them. On-demand mode adds an orchestrator layer: it watches for queued sessions, starts a runner as sessions arrive, and shuts them down when work finishes. On-demand keeps your capacity tracking demand rather than sitting idle.
Each session gets an isolated checkout, so one developer’s agent can’t accidentally touch another’s workspace.
What it solves
The obvious case is internal tooling. If your codebase needs a proprietary compiler, a licensed static analysis tool, or a database that only exists inside your VPN, Anthropic’s hosted runners couldn’t reach it. A self-hosted runner sitting inside your infrastructure can.
Compliance and data residency are the other drivers. Some organizations can’t send source code to a third-party server at all — regulatory requirements, audit constraints, or just internal policy. The self-hosted runner means the code stays put while the agent still works.
It’s also distinct from Remote Control, which ties Claude Code to an individual developer’s machine. Self-hosted environments are shared infrastructure: the platform team sets them up, and anyone in the organization can use them.
How to get it
The feature is off by default. To enable it, an admin with Claude Team or Enterprise access goes through the normal organization settings to provision and register runners. Anthropic published a quickstart guide in the Claude Code docs with the full setup steps.
One limitation worth noting: self-hosted environments aren’t available for organizations on Zero Data Retention plans, presumably because ZDR arrangements have their own audit requirements around how sessions are handled.
Context
This is the third infrastructure-level announcement Anthropic has shipped in the first week of August 2026, following the Inference Hooks DLP integration (which routes employee prompts through security tools before Claude processes them) and the Claude apps gateway for Amazon Bedrock and Google Cloud. The pattern suggests Anthropic is pushing hard on enterprise deployability, not just model capability, as the main product motion right now.
Sources: Anthropic blog · Claude Code docs · Unite.AI