Skip to content
NewsProduct launch

Anthropic opens self-hosted Claude Code runner beta, keeping source on customer infrastructure

· by Pondero Newsdesk

The short version

Anthropic launched a public beta of self-hosted environments for Claude Code on August 6, 2026, letting Team and Enterprise customers run cloud sessions on their own compute while source code stays on-premises.

Anthropic opens self-hosted Claude Code runner beta, keeping source on customer infrastructure

Until now, enterprises with hard data-residency requirements could not use Claude Code cloud sessions without source code passing through Anthropic-hosted infrastructure. That changed on August 6, 2026, when Anthropic opened a public beta of self-hosted environments for Claude Code, per the company's product blog. Organizations can now run those sessions on their own compute, with source code, build artifacts, and secrets staying on machines the organization controls. Only the conversation (prompts, tool results, and the session transcript) leaves the network for Anthropic's inference API.

What it is and how it works

The feature is available to organizations on Claude Team or Enterprise plans and is off by default. Organizations using Anthropic's zero-data-retention setting cannot use it.

Setup works through named environments in the claude.ai admin console. Each environment groups one or more runners: long-lived processes deployed on hosts inside the organization's network, working the same way as self-hosted CI runners. Runners come in two modes. Fixed mode keeps a static count running and distributes sessions across them. On-demand mode starts runners as sessions arrive and stops them when work finishes, so capacity tracks actual demand without keeping idle machines warm.

Every connection between a runner and Anthropic is outbound HTTPS to api.anthropic.com. No connection is ever initiated into the customer network; the announcement notes corporate egress proxies are supported. Model inference runs through Anthropic's own API; the beta does not support routing through Amazon Bedrock, Google Cloud, or Microsoft Foundry.

George Jacob, a senior engineering manager at Faire (a preview-program participant), said in the announcement: "Self-hosted environments let us integrate Claude Code into our existing development workflows while maintaining our security and operational controls. This setup means Claude can generate PRs, help fix CI issues, and respond to developer workflow events, with compute that can scale based on demand."

The documentation draws a clear line between this feature and Remote Control, an existing option that lets developers pick up a session running on their own machine from a phone or browser. Remote Control sessions end when the originating machine goes offline and belong to the individual user. The shared runner pool a platform team operates is what sets self-hosted environments apart: those sessions are available to any user in the organization.

Billing is unchanged. Usage counts against the organization's Claude Code plan the same as hosted sessions.

Why it matters

The three reasons preview participants adopted self-hosting are concrete, per Anthropic's blog: sessions can reach internal services and registries without public internet exposure; teams can pre-install compilers, SDKs, and internal CLIs so every session starts ready to build; and source code stays on infrastructure the organization controls for compliance purposes.

Most enterprises should stay on hosted. Real staffing is required for the self-hosted path, per Anthropic: a platform or developer-experience team must own setup, runner-image maintenance, and orchestrator operation. This is not a lightweight add-on.

For compliance and security teams that blocked cloud AI coding tools on data-residency grounds, the public beta gives a concrete path to evaluate Claude Code without that objection. Compliance-blocked teams that already run a platform or DevEx function should request beta access now. Teams without dedicated infra staffing should stay on hosted; Anthropic is explicit that the self-hosted path requires ongoing runner-image maintenance and orchestrator ownership.

What to watch next

Feedback goes through GitHub or through an Anthropic account team, per the documentation. The beta does not yet route sessions started from Claude Tag, Claude Security, or Code Review to self-hosted environments; support for those surfaces is coming separately.

Sources