Anthropic has opened a public beta of self-hosted environments for Claude Code, moving the coding agent’s cloud sessions off Anthropic’s infrastructure and onto servers inside the customer’s own network, the company announced on August 6, 2026. The option is available to organizations on Claude Team and Enterprise plans and is switched off by default.
A session started from the web, mobile, desktop, the terminal, or a scheduled routine runs inside the organization’s network, next to its internal services, toolchains, and security controls. Anthropic is steering most enterprises toward its hosted offering, which needs no infrastructure to run, and positioning self-hosting for teams whose network, tooling, or compliance requirements call for keeping agent execution on infrastructure they control.
The data split is specific. Repository checkouts, build artifacts, secrets, and any files a session creates or modifies stay on machines the organization provisions. The conversation itself, including prompts, responses, and tool results, goes to Anthropic’s API for model inference, and Anthropic stores the session transcript so a session can be picked up from any surface.
Anthropic said organizations in its preview program adopted the option for three reasons: sessions can reach internal services, databases, and registries without exposing them to the public internet; teams can pre-install compilers, SDKs, and internal command-line tools so every session starts ready to build; and source code and build artifacts remain in-house for compliance.
George Jacob, a senior engineering manager at Faire, 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 announcement distinguishes self-hosted environments from Remote Control, an existing feature that lets developers continue a session running on their own machine from a phone or browser. Remote Control sessions end when that machine stops and belong to the user who started them; self-hosted environments run on shared infrastructure a platform team operates, open to any user in the organization.
The launch extends a run of enterprise and infrastructure moves from Anthropic in the first week of August 2026. On August 5, 2026, the company put inline data loss prevention inside Claude Enterprise, and it also confirmed it is building an in-house silicon team for Claude.
How self-hosted environments work
An organization creates named environments in its claude.ai admin settings, each grouping a set of runners: long-lived processes, deployed on hosts inside the network, that work the same way as self-hosted CI runners. When a developer starts a cloud session and picks the environment, Anthropic’s control plane places the session on the environment’s queue. A runner with free capacity claims it, clones the chosen repository from GitHub using credentials the deployment provides, and starts a Claude Code process on the host to run the task, the documentation explains.
Runners come in two modes. Fixed mode keeps a set number running and distributes sessions across them; on-demand mode uses an orchestrator that starts runners as sessions arrive and stops them when the work finishes, so capacity tracks demand. A runner serves one user at a time, locking to that user’s account when it claims its first session, which keeps checked-out code from mixing between users. If a runner stops polling for about 60 seconds, the server requeues the session to another runner.
Every connection between the environment and Anthropic is outbound HTTPS to api.anthropic.com: queue polling, the session’s event stream, and model inference. Anthropic never connects into the customer’s network, and corporate egress proxies are supported.
The beta’s stated limits
The beta is limited to Claude Team and Enterprise organizations, and an owner or admin must enable it from the cloud-environments admin page, which requires Claude Code on the web to be turned on for the organization. Organizations using Anthropic’s zero-data-retention setting cannot use it.
Model inference stays with Anthropic. Sessions authenticate with Anthropic-issued, session-scoped tokens, and inference cannot be routed through Amazon Bedrock, Google Cloud’s Agent Platform, Microsoft (MSFT ) Foundry, or an LLM gateway. Sessions started from Claude Tag, Claude Security, and Code Review do not route to self-hosted environments yet; the documentation says support for those surfaces follows separately. Repository checkouts come from GitHub.
Billing is unchanged from hosted sessions, with usage counting against the organization’s Claude Code plan the same way. What changes hands is the operations: the organization builds and maintains the runner image, operates the fleet, and controls its network, and Anthropic says to plan on a platform or developer-experience team owning setup and ongoing maintenance. The path in runs through the admin toggle and a quickstart that covers installing Claude Code, creating an environment, starting a runner, and routing a first session.
