Now loading...
An innovative approach to infrastructure management has emerged with the introduction of self-hosted environments by Anthropic for organizations using its Claude platform. These environments prioritize data security by ensuring that all repository checkouts, build artifacts, secrets, and modified files remain on the user-provisioned infrastructure. While the conversation prompts, responses, and tool results—including code interpreted by Claude—are sent to Anthropic for inference, the session logs are retained to allow for continuity across devices.
Deploying self-hosted environments involves using a series of runners, which are persistent processes that manage sessions. Each session initiates a Claude Code process, with the runners operating in either fixed or on-demand modes. In fixed mode, a predetermined number of runners operate continuously, distributing sessions among them. Conversely, in on-demand mode, an orchestrator monitors session queues, activating a runner as new sessions arrive and shutting it down once tasks are completed, enabling efficient resource allocation based on demand.
Each runner can handle multiple sessions but maintains isolation between them, ensuring that developers and accounts do not interfere with one another’s work. Regardless of where a session is initiated within the environment, the setup remains consistent across the board, enhancing user experience and productivity.
It is critical to note that self-hosted environments operate differently than the Remote Control feature offered by Anthropic. Remote Control allows developers to continue sessions from their devices remotely, but these sessions conclude when the original machine ceases operation. On the other hand, self-hosted environments function on shared infrastructure overseen by the platform teams, making them accessible to any user within the organization.
Currently, self-hosted environments are in public beta, available for organizations subscribed to the Claude Team and Enterprise plans, though they are turned off by default for entities using Zero Data Retention (ZDR). It is recommended that a dedicated team manage the setup and ongoing operations, including the development and maintenance of the runner image and management of the orchestrator in on-demand mode. For more information about this offering, interested parties can consult the documentation and provide feedback through GitHub or their Anthropic account representative.
