Anthropic shipped a feature today that removes one of the most common reasons enterprise teams hesitate to adopt AI coding tools: the requirement to send your code to someone else’s servers.

Claude Code v2.1.224, released August 7, 2026, adds self-hosted environments. A single command, claude self-hosted-runner, turns your own machines or containers into the compute layer where Claude Code sessions execute. Web, mobile, desktop, and scheduled routine sessions can all run on infrastructure your team provisions and controls, rather than on Anthropic-managed cloud.

The feature is available on Team and Enterprise plans and is disabled by default.

Why This Changes the Conversation

Until now, every Claude Code session meant your repositories, build artifacts, and any secrets the session touched were processed on Anthropic’s infrastructure. For most developer teams building consumer apps or working on open-source projects, that has been an acceptable trade-off. For enterprises in regulated industries, it has been a blocker.

Financial services firms, healthcare organizations, government contractors, and anyone operating under strict data residency requirements have had limited options when evaluating AI coding tools. Either accept the data handling implications of cloud-hosted sessions, or run locally and give up the session portability that makes tools like Claude Code worth using across a whole team.

Self-hosted environments are a direct answer to that constraint. Your repository checkouts, build outputs, secrets, and any files a session creates or modifies stay on machines your organization provisions. Anthropic’s infrastructure handles the authentication and session routing, but the actual computation runs on your side of the boundary.

How It Works

The runner architecture is straightforward. You install the runner software on one or more machines or inside containers in your environment. Those machines register with the Claude Code backend and listen for sessions assigned to them.

Two modes are available:

Fixed runners keep a set number of machines running continuously. Sessions are distributed across them as they arrive. This suits teams with consistent, predictable AI coding workloads.

On-demand runners use an orchestrator that watches for queued sessions, starts a runner when work arrives, and stops it when finished. This is more cost-efficient for teams with variable or bursty usage patterns, and makes it straightforward to wire Claude Code into existing CI/CD infrastructure that already spins machines up and down.

Sessions launched from the web app, mobile, desktop client, or via scheduled routines can all target a self-hosted runner. From the end user’s perspective, the experience is identical to a cloud session. From the IT team’s perspective, the traffic never leaves the internal network.

Cross-Session Messaging Also Ships Today

Alongside the runner update, v2.1.224 adds cross-session messaging via two new tools: SendMessage and ListAgents. These let Claude Code sessions communicate with each other across machines.

The practical use case is multi-agent workflows. A long-running session handling a large refactoring task can now coordinate with shorter-lived sessions handling subtasks on other machines. For teams already running Claude Code routines at scale, this fills in a gap that required workarounds before.

What This Means for Business

For IT and security leaders evaluating AI coding tools, self-hosted environments significantly change the risk calculus. The core objection to cloud-hosted AI development, which is that your proprietary code travels outside your control, can now be resolved without abandoning the tool.

For data and engineering teams, the practical benefit is access to the full Claude Code feature set, including web and mobile sessions, scheduled routines, and the agent tools that make serious automation possible, without compromising on where the work actually runs.

This is particularly relevant for organizations building AI workflows on sensitive data pipelines. A team running data transformations, model training pipelines, or internal analytics tooling that touches customer data now has a path to AI-assisted development that stays inside their security perimeter.

The move also signals where Anthropic is positioning Claude Code for the enterprise segment. The previous model, where every session ran on shared cloud infrastructure, was a product built for developers. A self-hosted runner option, properly integrated with existing security controls and compliance posture, is a product built for enterprise IT departments.

If you are assessing AI development tools for your data or engineering team and security requirements have been the sticking point, this is worth evaluating. If you want help thinking through how AI-assisted development fits your team’s workflow, that is the kind of conversation we have through Omni Advisory.