Claude Code on the web vs your own cloud computer in 2026: what's the difference?

Claude Code on the web, which Anthropic's docs call cloud sessions, runs each Claude Code session in a fresh Anthropic-managed VM. The VM clones your GitHub repo and hands work back as a branch or pull request, with no separate charge beyond your Claude plan. Your own cloud computer, a VPS or a Sanbox Computer, is a machine you keep: its disk, installed tools and running services stay between sessions, and it can run other agents and model providers too. If your work is "repo in, pull request out" with Claude Code, Anthropic's option is usually enough. If the agent needs a machine that remembers more than the repo, or must run in EU data centres, use your own.

What is Claude Code on the web?

A cloud session is a Claude Code session that runs on Anthropic's infrastructure instead of your machine. You start one from claude.ai/code, the Code tab in the Claude mobile app, the Desktop app or claude --cloud in a terminal. It keeps running after you close your laptop. All facts below are from code.claude.com, checked 2026-11-06.

• Plans. Pro, Max, Ultra and Team, and Enterprise users with premium seats or Chat + Claude Code seats. Usage counts toward your plan's limits, and there's no separate charge for the VM.
• The machine. Each session gets a fresh Ubuntu 24.04 VM with common toolchains, Docker, PostgreSQL 16 and Redis 7.0 preinstalled. Resource ceilings are about 4 vCPUs, 16 GB of RAM and 30 GB of disk.
• Your code. Sessions clone from GitHub. Repos on GitLab or Bitbucket can be uploaded as a bundle, but the session can't push back to them.
• What carries over. Anything committed to the repo, such as CLAUDE.md and .claude/settings.json. Your user-level ~/.claude files, local MCP servers and installed tools don't.
• Setup. A setup script runs before Claude starts. If it finishes in about five minutes, Anthropic caches the resulting filesystem for later sessions and rebuilds it after about seven days.
• Idle sessions. After a few minutes without activity the VM pauses with its files saved, and a paused VM can later be reclaimed. Reopening a reclaimed session gives you a fresh VM with your conversation history, but shell commands and background work that were running don't come back.
• Network. Each environment has an access level: None, Trusted (the default: package registries, GitHub, cloud SDKs), Full or Custom. GitHub credentials stay on Anthropic's servers, and network secrets are attached to requests after they leave the VM.

Cloud sessions aren't available with Zero Data Retention, the HIPAA configuration or customer-managed encryption keys, and they need an Anthropic account rather than Amazon Bedrock or Google Cloud. An organisation IP allowlist makes Anthropic-hosted sessions fail unless Anthropic exempts them.

What does "your own cloud computer" mean?

A machine in the cloud that you or your agent keep, rather than one created per session. Two common forms:

• A VPS you run. You install Claude Code, keep it alive in tmux and patch the server yourself. Anthropic's Remote Control lets you steer that session from claude.ai/code or the Claude app.
• A Sanbox Computer. A Firecracker microVM with its own kernel and private disk, in EU data centres. OpenCode Computers stay on until you stop them, and a stop keeps /workspace and installed packages. Sanbox has no Claude Code template, so you install it yourself:

sanbox run --template <opencode-computer-template-id> --json
# The image has Node 22; install once into the system path:
sanbox exec <run-id> --root -- npm install -g @anthropic-ai/claude-code
# then work as the normal user:
sanbox ssh <run-id>

The template's network policy must allow registry.npmjs.org and api.anthropic.com, plus any other hosts Claude Code needs. Run it inside zellij so it survives a disconnect, or use the console's browser terminal, which keeps sessions open after you close the tab. SSH and the browser terminal use different home directories, so a Claude Code login made in one isn't visible in the other. The 24/7 Claude Code guide has the full recipe.

How do they compare?

When is Anthropic's option enough?

• Your task starts from a GitHub repo and ends in a branch or pull request.
• You want to fan out several tasks at once and review the diffs later.
• You want Claude to watch a pull request and fix CI failures or review comments (auto-fix).
• You don't want to pay for or maintain another machine.
• Your setup can be scripted in about five minutes, and nothing needs to keep running between sessions.

When do you want your own cloud computer?

• The agent needs state beyond the repo: installed tools, caches or a database with data in it.
• Services must stay up: a dev server, a queue worker or a long test suite that shouldn't stop when you look away.
• You use another agent or provider: Codex, OpenCode or Hermes, or Claude through Bedrock or Google Cloud.
• Your repo isn't on GitHub and the agent must push back to it.
• The machine must run in the EU, under a DPA with an EU company. Sanbox stores and processes Computers, disks, snapshots, logs and run records in EU data centres, under an Article 28 DPA with Sanlabs GmbH. Model calls still go to the provider you configure.

What about Remote Control and self-hosted environments?

Remote Control connects claude.ai/code or the Claude app to a Claude Code session running on your own machine. It is available on Pro, Max, Ultra, Team and Enterprise plans, not with API keys. Your computer has to stay on and the claude process has to keep running. The docs suggest tmux or screen when the machine is remote.

Self-hosted environments are in public beta on Team and Enterprise plans. Cloud sessions run on runners inside your network, but the conversation, including prompts, responses and tool results, still goes to api.anthropic.com. Sanbox is not a self-hosted environment runner and doesn't integrate with them. A Sanbox Computer is a separate machine where you run the agent yourself.

Where Claude Code on the web is stronger

• No extra bill and no setup. It comes with your Claude plan and starts from a browser or phone.
• GitHub integration: cloning, branches, pull requests and auto-fix on pull requests.
• Parallel sessions from one command each, plus teleporting a session into your terminal.
• More disk out of the box. About 30 GB per session, against 1 to 8 GB self-serve on a Sanbox Computer.
• Scheduled cloud tasks are self-serve. On Sanbox, scheduled runs are enabled per organisation on request.
• Maintained by Anthropic alongside Claude Code itself.

When not to use Sanbox

• Claude Code on the web already covers your workflow. Adding a Computer only adds a bill.
• You need GPUs or WebSockets into the machine.
• You want a ready-made Claude Code setup. On Sanbox you install and sign in to Claude Code yourself.

For a business use case, talk to us. Our forward-deployed engineers can build it with you.

Related

• How do I keep Claude Code running 24/7 and control it from my phone?
• How do I run Claude Code on a cloud computer that keeps its files?
• How do I log in to Claude Code or Codex on a headless server?
• Docker Sandboxes vs a persistent cloud computer

Sources

• Anthropic, Use Claude Code in the cloud (plans, surfaces, GitHub, environment expiry, limitations), checked 2026-11-06
• Anthropic, Configure cloud environments (installed tools, resource and time limits, network levels, network secrets, caching), checked 2026-11-06
• Anthropic, Remote Control (plans, local process, tmux), checked 2026-11-06
• Anthropic, Self-hosted environments (public beta, Team and Enterprise, what stays on your infrastructure), checked 2026-11-06
• Anthropic, Data usage (cloud execution data flow), checked 2026-11-06