By ByteBot
Anthropic opened the public beta of self-hosted environments for Claude Code on August 6, 2026. Organizations on Team and Enterprise plans can now run Claude Code sessions on their own machines and containers, keeping repository checkouts, build artifacts, secrets, and internal toolchain access inside their network rather than routing through Anthropic’s managed infrastructure. The feature is available today, off by default, and requires an Owner or admin to enable it in the claude.ai Cloud environments admin page.
How It Works
The claude self-hosted-runner command deploys long-lived runner processes on your hosts. When a developer starts a Claude Code session — from the web app, mobile, desktop client, terminal, or a scheduled routine — Anthropic’s control plane queues it in your environment. A runner claims the session, clones the repository the developer selected, and starts a Claude Code process on your infrastructure. Sessions from every surface route to the same environment, so you configure once and it works everywhere your team works.
All connections between your runners and Anthropic are outbound HTTPS to api.anthropic.com. Anthropic never initiates a connection into your network. Corporate egress proxies are supported.
What “Self-Hosted” Actually Means Here
Before you brief your security team on this as a zero-cloud solution, read the fine print. Repository checkouts, build artifacts, secrets, and any files a session creates stay on your infrastructure. The conversation itself — prompts, responses, and tool call results — still routes to Anthropic’s API for model inference. Your code repo stays on-prem; the chat does not.
For most teams, this is acceptable. Internal APIs and credentials never leave the perimeter. Engineers can point Claude Code at internal databases, registries, and services that are not reachable from the public internet. For compliance-driven teams in healthcare, finance, or defense who need AI coding tooling but could not greenlight cloud-only agents, this clears a real blocker. For teams that need full conversation-level air-gapping — the kind where even prompts cannot leave — this is not there yet. Organizations with Zero Data Retention (ZDR) agreements are explicitly excluded from the beta.
Fixed or On-Demand: Pick Your Runner Mode
Runners come in two modes.
Fixed keeps a set number of runners alive at all times. Sessions are distributed across them as they arrive. Good for teams with predictable, steady workloads and platform engineers who prefer simple operations.
On-demand has an orchestrator watch for queued sessions, spin up a runner when one arrives, and shut it down when work is done. Capacity tracks demand automatically. The tradeoff: your platform or developer productivity team owns the orchestrator’s setup and ongoing maintenance. If that sounds like more ops overhead than you want right now, start with fixed mode.
Each session runs in its own checkout for isolation even when runners are shared, so parallel sessions do not bleed into each other.
Prerequisites and What to Check
To use self-hosted environments, your organization needs to be on Claude Team or Enterprise. Claude Code on the web must be enabled first — self-hosted environments live under that feature flag. An Owner or admin then turns on Allow self-hosted environments in the Cloud environments admin page.
Plan on a platform, developer experience, or developer productivity team owning runner image builds, updates, and orchestrator operations. This is not a one-click toggle and walk away. Your team picks it up and maintains it, which is the cost of getting network-level isolation.
The Larger Picture
This is Anthropic’s third enterprise announcement in three days. On August 5, the company shipped Inference Hooks, which routes every employee prompt through your security server for an allow-or-deny verdict before the model sees it. On the same day, Anthropic confirmed it is building an in-house silicon team to own its compute stack. Self-hosted runner environments are the third move. Taken together, Anthropic is systematically closing the gap between “best AI coding tool” and “enterprise-deployable AI coding tool.”
Claude Code is the most-used AI coding tool heading into August 2026. Adding infrastructure control brings it into regulated industries that Cursor, GitHub Copilot, and other cloud-only tools could not reach. Organizations already self-hosting open-weight alternatives specifically for compliance reasons now have a strong reason to look again.
What to Do Now
If your organization is on Team or Enterprise, check whether Claude Code for the web is enabled, then have an admin enable self-hosted environments from the Cloud environments admin page. Anthropic’s quickstart guide walks through standing up one runner on a single host — the smallest working configuration — to verify the setup before you commit to a full rollout.
If you are on a ZDR plan, self-hosted environments are not yet available. Watch the Claude Code changelog for updates on ZDR support and on-premises inference options.
