# Claude Code's cloud sessions or your own machine: which to use when > What Claude Code's cloud sessions do well, where they stop, and when a Linux machine of your own fits better, checked against Anthropic's docs. *2026-11-02 ยท claude-code, remote-dev, costs, workflows* Claude Code can run on Anthropic's computers instead of yours. Anthropic calls these cloud sessions, and in the browser the feature is also known as Claude Code on the web. You give it a task from your phone or laptop, and it keeps working after you close the lid. That covers a lot of what people want a cloud machine for, at no extra charge. So when is a machine of your own still worth having? This comparison comes from Anthropic's [cloud session docs](https://code.claude.com/docs/en/claude-code-on-the-web) and [environment docs](https://code.claude.com/docs/en/cloud-environments) as they read on 2 November 2026. Limits like these change, so check them before you plan around one. ## What a cloud session is Each session gets a fresh virtual machine that Anthropic manages: Ubuntu 24.04 on x86_64, with your repository cloned and common toolchains installed. That includes Node, Python, Go, Rust, Java, Docker with Compose, PostgreSQL 16 and Redis 7.0. You can start one from claude.ai/code, the Code tab in the Claude app, the Desktop app, or a terminal: ```bash claude --cloud "Fix the flaky test in auth.spec.ts" ``` Cloud sessions come with the Pro, Max, Ultra and Team plans, and with Enterprise premium or Chat + Claude Code seats. Each one runs in an environment: a saved setting for network access, environment variables and a setup script. ## What it does well Simon Willison put the main appeal plainly when it launched: "the benefit of Claude Code for web is entirely in its convenience" ([his post](https://simonwillison.net/2025/Oct/20/claude-code-for-web/)). The same prompt on his laptop would likely have ended the same way. What you gain is everything around the agent. - **Nothing extra to run or pay for.** Sessions count toward your plan's usage like any other Claude Code use, with no separate charge for the VM. There's no machine to start, patch, or forget to stop. - **Starting from anywhere.** Ben Borgers [treats it](https://ben.page/claude-code-web) as a queue for small changes across his projects, reviewed whenever he's next at a computer, and uses the Claude app to send it questions while he's away from one. - **Parallel work.** Every `claude --cloud` starts its own session on its own VM. Your plan's usage is the limit, and parallel tasks use it up faster. - **Credentials kept off the machine.** Your GitHub credentials never enter the VM. Git and the built-in GitHub tools go through a proxy that adds them on Anthropic's side, and keys for other services can be stored as network secrets, attached to requests after they leave the session. On a machine of your own, a token you sign in with sits on its disk. - **A closed network by default.** The default Trusted level allows an allowlist of package registries, GitHub and cloud SDKs. None blocks the session's own outbound traffic, though GitHub, through its proxy, and Anthropic's API stay reachable. Willison considered no network access the safest choice and was uneasy about how much the Trusted list lets through. - **A cached setup.** If your setup script finishes in about five minutes, Anthropic snapshots the result, and later sessions start with your dependencies and Docker images already on disk. - **Pull request follow-up.** With the Claude GitHub App installed, auto-fix watches a pull request for failing CI and review comments, and pushes a fix when one is clear. ## Where it stops - **GitHub for cloning and pushing.** Cloning and pull requests require GitHub. A repository on another host can go up as a bundle with `CCR_FORCE_BUNDLE=1` (under 100 MB, untracked files left out), but the session can't push back to that remote. The GitHub proxy also refuses GraphQL, so `gh pr` and `gh issue` fail inside a session; Claude uses its built-in GitHub tools and `gh api` instead. - **No shell of your own.** The docs are direct: you don't get a shell into the VM, and Claude runs every command for you. There's no SSH, the Desktop app shows a cloud session's files read-only, and the docs describe no way to open a dev server it runs in your own browser. - **A size ceiling.** Roughly 4 vCPUs, 16 GB of RAM and 30 GB of disk, and the VM may stop tasks that need much more memory. For bigger work, the docs themselves suggest running Claude Code on your own hardware with Remote Control. - **Time limits.** A command gets 2 minutes by default (up to 10), then moves to the background with 30 more minutes before it's stopped. A command Claude starts in the background gets 30 minutes, or up to 2 hours if Claude asks. Environment variables raise these limits. - **Idle means paused, and later gone.** After a few minutes without activity, the VM pauses with its files saved, and a paused VM can later be reclaimed. Reopening the session then brings back the conversation but not the shell commands, subagents or scheduled tasks that were running. The setup cache keeps files, not processes, so a database or a Compose stack starts again each session. - **Your own setup stays behind.** Only what's committed to the repository comes along. Your user-level `CLAUDE.md`, user skills and MCP servers added at user or local scope stay on your computer, and sign-ins that need a browser, such as AWS SSO, aren't supported. ## What your own machine adds A Linux machine of your own, whether a VPS, a computer at home or a cloud VM, turns most of those around: - **A shell and your tools.** SSH in, attach to the agent's tmux session, open the code in an editor, fix something by hand. - **Things that stay.** The disk keeps installs, databases and worktrees between tasks, and a dev server can run all afternoon. - **The size the work needs**, and a way to see the app: forward its port over SSH, or share a link. - **Any git host**, your own dotfiles and MCP servers, and whatever you choose to sign in to. It has costs of its own. You set it up, keep it patched and pay for it while it runs. What it can reach on the network is up to you, and every sign-in you make on it is stored on its disk, inside the agent's reach. ## Using both You can start a task in the cloud from your phone, then pull it onto your machine to finish with your own tools: ```bash claude --teleport ``` Run it in a checkout of the same repository with no uncommitted changes, signed in to the same claude.ai account. It fetches the session's branch and loads the conversation. Your terminal gets its own copy: new work there stays local and doesn't appear in the cloud session. To keep steering from your phone, start `/remote-control` in it. On a remote machine, make sure git can fetch without asking for a password, since teleport's fetch never waits for input. Going the other way, the CLI can't push a terminal session to the cloud, but the Desktop app can send a local session there. The Desktop app can also run sessions on your own machine over SSH, installing Claude Code there the first time you connect. ## What each costs A cloud session costs nothing beyond your plan, but it draws on the same usage limits as everything else you do with Claude, faster when sessions run in parallel. A machine of your own adds its price to that usage: a monthly VPS, hardware at home, or a cloud VM billed by the hour. ## Which to use when Use a cloud session for well-defined tasks on a GitHub repository, work you start from your phone, several tasks at once, and work where you'd rather keep credentials and the open internet away from the agent. Use your own machine for long-running services, builds bigger than 4 vCPUs and 16 GB, repositories outside GitHub, work you want to watch or steer in a terminal or editor, apps you need to look at, and setups that depend on your own tools and sign-ins. ## Where shedix fits shedix is the second kind, billed by the hour: a Linux machine of your own on Google Cloud, running Ubuntu 24.04 with sudo, Docker, Claude Code and Codex. An M has 4 vCPUs and 8 GB for $0.16/hr while it runs, so 8 hours on each of 22 workdays comes to $28.16 a month. An L doubles that for $0.32/hr, and an XL, from the CLI or API, doubles it again. A machine sleeps after an hour unused and keeps its disk. To bring a cloud session onto one, clone the repository there, sign in to Claude Code in the machine's own terminal, and connect GitHub from the machine's page so git can fetch without a prompt. Then run `claude --teleport` in the checkout. Unlike a cloud session, a shedix machine's outbound internet is open, and the sign-ins you make stay on it. [Close the laptop](/use-cases/close-the-laptop) shows a session moving onto one, and [Keep Claude Code running after you close your laptop](/blog/keep-claude-code-running-after-closing-laptop) covers the tmux side.