Run Claude Code Sessions on Your Own Machines: v2.1.224 Self-Hosted Runner

Claude Code v2.1.224 (Aug 7) lets Team/Enterprise teams run cloud sessions on their own hardware with a single command — keeping code, secrets, and tool execution on infrastructure you control.

Before this release, Claude Code cloud sessions (from the web, mobile, or desktop app) ran on Anthropic's compute. Your code was transferred to Anthropic-managed infrastructure for the duration of the session. For many teams in regulated industries or behind network perimeters, that was a blocker.

What changed

v2.1.224 introduces claude self-hosted-runner, which turns your own machine or container into the compute layer for Claude Code cloud sessions.

# On the machine you want Claude Code to run on:
claude self-hosted-runner

# That's it — Claude Code web/mobile/desktop sessions for your org
# now execute on this machine instead of Anthropic's servers

The command is available on Team and Enterprise plans. Sessions launched from any client (web, desktop, mobile) are routed to your self-hosted runner instead of Anthropic's compute.

Three reasons to use it

Network access. Your Claude Code sessions run inside your network. The agent can reach internal APIs, databases, artifact registries, and dev services without exposing them to the internet or adding VPN complexity.

Toolchain control. Pre-install your compilers, SDKs, proprietary CLIs, and custom tools on the runner host. The agent finds them just as it would on a developer's local machine.

Compliance. Source code, environment variables, and intermediate artifacts never leave infrastructure you control. For teams in HIPAA, FedRAMP, or SOC 2 environments, this closes the biggest gap in cloud-based AI coding workflows.

Cross-session messaging (same release)

v2.1.224 also ships cross-session messaging: agents in separate sessions can now send messages to each other by name. Combined with self-hosted runners, this enables multi-agent pipelines that run entirely on your own infra — one runner handles orchestration while others handle sub-tasks, all within your network perimeter.

# In one session, send a message to another named session:
# (from Claude Code's perspective, via the ListAgents / SendMessage tools)
# Sessions on the same runner can coordinate without API round-trips

When to use self-hosted runners

• Your repo contains secrets that can't leave your network
• The agent needs to call internal microservices or hit a private artifact registry
• Your organization's security review prohibits code execution on third-party SaaS compute
• You need deterministic toolchain versions across all agent sessions

For teams without these constraints, Anthropic-managed compute is simpler and still the right default.

Sources: Claude Code v2.1.224 release notes — GitHub · Claude Code self-hosted runner overview — clauding.de · Claude Code major updates v2.1.224 — dev.classmethod.jp · Claude Code changelog — code.claude.com