CloudCodeTree LogoCloudCodeTree
AI NewsTutorialsAbout
CloudCodeTree Logo
CloudCodeTree
  • AI News
  • Tutorials
  • About
← Back to AI News
Claude Code Self-Hosted Environments Are Now in Public Beta: Run Agent Sessions on Your Own Compute

Photo: panumas nikhomkhai / Pexels

Claude Code Self-Hosted Environments Are Now in Public Beta: Run Agent Sessions on Your Own Compute

Chris Harper

2 min read

Aug 7, 2026 · 12:05 UTC

AI
Workflow
Claude Code
Best Practices

Claude Code's new self-hosted environments (public beta, Team/Enterprise) run agent sessions on your own compute — inside your firewall, next to internal APIs, keeping source code and artifacts off Anthropic's servers.

Anthropic's cloud infrastructure covers most workloads, but regulated industries, internal-API-dependent codebases, and strict compliance requirements need agent execution on controlled compute. Self-hosted environments solve this while keeping the full Claude Code cloud UI.

How it works

You deploy runners — long-lived processes that pick up session requests from Claude Code clients (web, mobile, desktop, routines) and start a local Claude Code process per session. Two modes:

  • Fixed: a set number of runners stay running; sessions distribute across them
  • On-demand: an orchestrator watches the queue and provisions runners as sessions arrive

Repository checkouts, build artifacts, secrets, and any files a session creates stay on your machines. The session UI runs through Anthropic's cloud; only the agent execution is local.

Set up (Team or Enterprise plan)

  1. Go to your organization's Claude settings → EnvironmentsSelf-hosted (beta)
  2. Deploy runners on your infrastructure (the docs cover Kubernetes and systemd deployments)
  3. Runners register with a token and start accepting sessions immediately

No changes to how developers start a session — they get the same web/desktop/mobile UI; the agent just executes on your hardware.

Also shipping in this update: cross-session messaging + spawn cap gone

Two more agent coordination improvements alongside self-hosted environments:

  • Cross-session messaging: ListAgents now discovers Claude Code sessions across machines (macOS and Linux); SendMessage delivers work between them without human relay. Background agents can hand off to other sessions directly.
  • Subagent spawn cap removed: the previous 200-subagent ceiling per session is gone. Long autonomous workflows no longer hit a hard limit — concurrency and depth limits still apply, but sustained multi-hour agent runs are no longer bounded by a count.

Sources: Self-hosted environments for Claude Code — Anthropic blog · Self-hosted environments docs — code.claude.com · Claude Code changelog Aug 2026 — Gradually