If you’ve been running Claude Code or OpenAI Codex on your local machine, you’ve probably hit that moment where the agent wants to do something heavy — train a model, spin up a container, run GPU-intensive tasks — and your laptop just stares back at you, fans screaming. [CloudRouter](https://cloudrouter.dev/) is a new tool that basically solves this by letting your AI coding agent spin up its own cloud VM or GPU instance on demand, do its thing, and tear it down when it’s done.

The idea is simple but the execution matters. You install CloudRouter as a skill for Claude Code (or hook it into Codex, Cursor, etc.), and from that point on, the agent can launch a remote sandbox straight from your local project directory. It uploads your files, runs whatever commands it needs on the remote machine, and syncs everything back. You get access to the environment through VS Code in the browser, a terminal, or even a VNC desktop if you need a full GUI. Docker is built in, file syncing is automatic, and GPU options range from T4s all the way up to H100s — you can even request multi-GPU setups like `–gpu H100:2`.

What caught my attention is how this showed up on [Hacker News as a Show HN post](https://news.ycombinator.com/item?id=47006393) on Feb 13 and quickly picked up 85+ points with some genuinely interesting discussion. People debated the architecture choices, asked about security (connections go through authenticated TLS WebSockets with ephemeral VMs, which is reasonable), and some questioned why you wouldn’t just use raw AWS or GCP CLIs. The answer, honestly, is convenience — CloudRouter wraps all the boring setup into a single workflow that an AI agent can drive without hand-holding.

The timing feels right. With the explosion of AI coding agents over the past few months, the biggest friction point has been resource isolation. You don’t want three parallel Claude Code sessions fighting over your local CPU and memory. CloudRouter gives each agent its own disposable environment, which is exactly the kind of infrastructure layer this ecosystem has been missing. Built in Go and distributed as npm packages, it’s MIT licensed and works across macOS, Linux, and Windows.

It’s still early — the team is working on bring-your-own-cloud support, and there are rough edges in the setup flow. But if you’re deep in the AI-assisted coding workflow and keep bumping into local resource limits, [CloudRouter](https://cloudrouter.dev/) is worth a look.