Anthropic launches self-hosted beta for Claude Code
Anthropic has launched a public beta of self-hosted environments for Claude Code, allowing enterprise teams to run the coding agent's sessions securely within their own private networks.

On August 6, 2026, Anthropic launched a public beta of self-hosted environments for Claude Code, allowing organizations to run the coding agent's cloud sessions on their own internal servers rather than Anthropic's infrastructure. Available as an opt-in feature for Claude Team and Enterprise plans, this setup keeps repository checkouts, build artifacts, secrets, and modified files entirely within the customer's network. However, the actual conversation transcripts, prompts, and tool results are still sent to Anthropic's API to handle model inference.
To set this up, administrators create named environments in their claude.ai settings to manage runners, which operate similarly to self-hosted continuous integration runners. These runners connect to Anthropic via outbound HTTPS to api.anthropic.com, meaning no inbound connections into the private network are required. The system supports two operational modes: a fixed mode that maintains a set number of active runners, and an on-demand mode that scales runners up or down based on active sessions. To prevent data mixing, each runner locks to a single user's account while active, and sessions are requeued to another runner if polling stops for approximately 60 seconds.
For development teams, this hybrid model allows Claude Code to securely access internal databases, compilers, SDKs, and private command-line tools without exposing them to the public internet. George Jacob, a senior engineering manager at Faire, noted that the setup lets them "integrate Claude Code into our existing development workflows" while keeping operational control. However, practitioners must note some limitations. Model inference cannot be routed through third-party platforms like Amazon Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry. Additionally, the beta does not yet support Claude Tag, Claude Security, or Code Review, and organizations using Anthropic's zero-data-retention setting are excluded.
This release follows other major enterprise updates from Anthropic, including the August 5, 2026, introduction of inline data loss prevention for Claude Enterprise and confirmation of an in-house silicon team. While billing remains identical to standard hosted sessions, customer platform teams must take on the responsibility of building, maintaining, and operating the runner images.
This is our own summary of reporting by Unite.AI



