---
title: "Claude Code in the cloud: one machine per task"
description: "A cloud session gives each task its own machine and branch: several jobs run side by side, survive a sleeping laptop, and come back as pull requests."
canonical: "https://me.consonance.fyi/en/blog/claude-code-in-the-cloud/"
lang: "en"
published: "2026-10-07"
---

# Claude Code in the cloud: one machine per task

> A cloud session gives each task its own machine and branch: several jobs run side by side, survive a sleeping laptop, and come back as pull requests.

By Minh-Toan Ta · 7 October 2026 · 5 min read

Picture three carpenters sharing a single workbench. One is sanding, one is gluing, one needs the vice the other two keep borrowing. Nobody is bad at their job; there is just one bench. Now give each of them a workshop of their own, with a copy of the plans, and ask them to bring back the finished piece for inspection. Same people, same work, no elbows.

That is, more or less, the difference between running Claude Code in a terminal on your laptop and running it in the cloud. I wrote a short sheet about it on [Consonance](https://consonance.fyi/en/mode-d-emploi/claude-code-in-the-cloud); this is the longer version, with the reasoning behind each line.

## The idea in one sentence

A cloud session runs Claude Code on a machine of its own, one per task, with your repository on a new branch: several tasks run in parallel without getting in each other's way, and the work comes back as a branch or a pull request.

## Why the laptop is the bottleneck

A local session depends on your laptop in three ways. It shares your working tree, so two sessions on one repository can edit the same files and fight over the same port. It runs with your credentials. And it stops when your computer sleeps or the Wi-Fi drops.

A cloud session removes all three. Each task gets a fresh virtual machine, your repository cloned onto a new branch, and your environment's setup already done. You can start one from claude.ai/code, the Claude mobile app, the Desktop app, your terminal or Slack, and follow it from the browser, the phone or Desktop. Close the lid: the session keeps going.

According to the field guide on claude.dev, cloud sessions come with Pro, Max, Team and Enterprise plans at no extra cost. There is no separate charge for the machine, and sessions draw on the same usage limits as the rest of Claude Code.

_One machine per task: each session clones the repository onto its own branch and ends with its own pull request._

## Three tasks, three machines, 87 seconds

The claude.dev guide backs this up with a real run. The sample repository, tidepool, is a small Node API with three ordinary problems: a test that failed about one run in four, API docs that no longer matched the code, and a logger that built its lines by concatenating strings. The author started three cloud sessions within 16 seconds of each other, one per problem. All three were done 87 seconds after the first one started.

The results are worth reading for what they say about method, not speed:

- **The flaky test.** Claude found a race in the cache, fixed the root cause in the code rather than the test, then ran the suite 40 times in a row: 0 failures. The prompt had asked for at least 30.
- **The docs.** Claude started the server, ran a curl against every endpoint and found five ways the old doc was wrong.
- **The logger.** Claude wrote the JSON logger and added five tests. Along the way it hit the same cache bug the first session was fixing. It reported it, and left it alone, because it was outside its task.

That last point is the one I keep coming back to. Isolation cuts both ways: the logger session could not see the fix happening next door. So split parallel tasks along file boundaries, merge the branches in a sensible order, and expect a session to report problems another one is already solving.

> **Ask for proof, not just a patch**
>
> The VM's CPU isn't yours, so be generous with verification. Write each task like a self-contained ticket: what is wrong, what done looks like, and the command that proves it. "Run the suite at least 30 times" is what turned a guess into 40 clean runs.

## What happens under the hood

Four details change how you work day to day:

- **Every task gets its own machine.** Sessions cannot touch each other's files or ports.
- **Your GitHub token never enters the VM.** A proxy holds it, and the session works with a scoped, short-lived credential instead.
- **The repository's Claude config comes along, your personal one doesn't.** `CLAUDE.md`, rules, skills, agents and commands travel with the repo; your `~/.claude` stays on your laptop. If Claude needs to know how to run the integration tests, the repository has to say so.
- **Idle VMs get reclaimed.** Reopen the session and you get a fresh machine with the conversation restored, so commit as you go.

## Local or cloud?

Cloud sessions don't replace local ones; most people use both. Stay local when the task needs something only your machine has: a database with real local data, a service behind a VPN, a GPU, a phone simulator, hardware on your desk. Stay local too for tight visual loops where you want to see each change in your own browser within seconds. For everything else, the cloud is a good candidate.

## Try it

**1. Start a task in the cloud from your terminal.** `claude --cloud` clones your GitHub remote at your current branch, so push your commits first.

```
claude --cloud "Fix the flaky test in auth.spec.ts"
```

To clear a small backlog, start one session per task, exactly as in the guide, and review the branches when they land:

```shell
claude --cloud "Fix the flaky test in auth.spec.ts"
claude --cloud "Update the API documentation"
claude --cloud "Refactor the logger to use structured output"
```

**2. Bring a finished session back into your terminal.** Teleport checks that you are in the same repository, fetches the session's branch, checks it out and loads the whole conversation. You need a clean working tree (it offers to stash), and the branch has to be pushed.

```
claude --teleport
```

**3. Or connect GitHub from the terminal** with your existing `gh` token, from inside Claude Code. On Team and Enterprise plans, an owner has to turn on Quick setup first.

```
/web-setup
```

> **Signing in with GitHub is not enough**
>
> Signing in tells Claude who you are. For private repositories you also need to install the Claude GitHub App on the account or organization that owns them, or they simply won't show up in the picker. Auto-fix and GitHub-triggered features depend on the App too.

## A few habits

One task, one session: small sessions are easier to review and cheaper to throw away. Read Claude's summary before the diff. And keep an eye on your limits: parallel sessions draw on your plan in parallel, so five sessions use it about five times as fast as one.

## Wrapping up

What I like about this model is that it turns a coding agent into something closer to a team workflow: tickets go out, branches come back, you review. The laptop stops being the place where everything has to happen and becomes the place where you decide what gets merged. Give each task its own workshop, ask for proof, and spend your attention on the pull requests.

*Independent write-up, not affiliated with Anthropic. Claude is a trademark of Anthropic, PBC. Every command and figure comes from the sources listed below.*

## Sources

- [Consonance · Claude Code in the cloud: one machine per task](https://consonance.fyi/en/mode-d-emploi/claude-code-in-the-cloud)
- [claude.dev · Claude Code in the cloud: a field guide to cloud sessions](https://claude.dev/blog/claude-code-in-the-cloud/)
- [Claude Code docs · Claude Code on the web](https://code.claude.com/docs/en/claude-code-on-the-web)
- [Claude Code docs · Cloud environments](https://code.claude.com/docs/en/cloud-environments)

---

- HTML: https://me.consonance.fyi/en/blog/claude-code-in-the-cloud/
- [Français](https://me.consonance.fyi/blog/claude-code-in-the-cloud.md) · [Tiếng Việt](https://me.consonance.fyi/vi/blog/claude-code-in-the-cloud.md)
- MCP: https://mcp.consonance.fyi
- Contact: tmtoan2601@gmail.com · https://www.linkedin.com/in/minhtoanta/