What Are Claude Code Cloud Sessions? How They Keep Running With Your Laptop Closed, How to Start, and How to Claim the $100 Pro / $250 Max Credit

On September 24, 2026 (around 6:23 a.m. Japan time, JST), the official Claude Code account @ClaudeDevs announced that cloud sessions are out of research preview and officially available. At the same time, existing paid subscribers are getting a one-time credit to try them ($100 on Pro, $250 on Max).

Cloud sessions run Claude Code in Anthropic's cloud instead of on your own computer. Your GitHub repository is cloned into a virtual machine in the cloud, where Claude edits the code, runs the tests and pushes a branch. Closing your laptop doesn't stop the work. You can also check on it and reply from your phone.

This article draws on the official Claude Code documentation (Use Claude Code in the cloud, Configure cloud environments and Get started with Claude Code in the cloud), the @ClaudeDevs posts and the full text of the credit offer terms to lay out what cloud sessions can do, what they can't, what they cost and how to claim the credit. Information is current as of September 24, 2026.

1. What cloud sessions are: it isn't your computer doing the work

The official documentation's definition is short: "a Claude Code session that runs on cloud infrastructure instead of on your machine." By default it runs on infrastructure Anthropic manages (if your organization has its own self-hosted environment, sessions can be routed there instead). claude.ai/code, which you open in a browser, is also known as "Claude Code on the web" and is one of the ways to start a cloud session.

When you send a task, this is what happens in the cloud.

Stage ④ is easy to misread. Pull requests aren't opened automatically: you create one yourself with "Create PR" in the diff view (as a full PR, as a draft, or by jumping to GitHub's compose page). The session doesn't close after the branch is pushed, so creating the PR and any follow-up fixes happen in the same conversation.

The session keeping going after you close the tab is by design. The official documentation puts it this way: closing the tab doesn't stop the session, which keeps running in the background until Claude finishes the current task and then idles. If it stays idle, though, the virtual machine is reclaimed. Your conversation history is kept, and reopening the session from claude.ai/code resumes it on a fresh VM, but any subagents or shell commands that were running at the time are not restored.

Cloud sessions are available on Pro, Max and Team, and for Enterprise users with premium seats or Chat + Claude Code seats. Organizations with Zero Data Retention enabled can't use them.

2. How they differ from local Claude Code and Remote Control

The official documentation says: "Claude Code behaves the same everywhere. What changes is where the session runs and whether your local configuration is available." There are two things to compare against: a local session you run in a terminal or IDE, and Remote Control, which lets you drive a local session from your phone or a browser.

The row that matters most is "your local config isn't used." Every cloud session starts from a fresh clone of the repository, so only what is committed to the repository comes along. Summarizing the lists in the official documentation:

• Carried over: the repository's CLAUDE.md, .claude/rules/, .claude/skills/, agents/ and commands/, and your organization's server-managed settings. In a session with a single repository, also the hooks and permission rules in .claude/settings.json and .mcp.json
• Not carried over: your own ~/.claude/CLAUDE.md and ~/.claude/skills/, MCP servers you added only locally with claude mcp add, plugins enabled only in your user settings, and authentication that needs a browser sign-in, such as AWS SSO

Permission modes differ too. In a cloud session you can choose Auto (a classifier reviews Claude's actions; it appears only when your organization allows it and the selected model supports it), Accept edits (changes are made and the branch is pushed without waiting for approval) or Plan (Claude proposes an approach and waits for your approval). There is no Manual mode, where you approve each action, and no Bypass permissions, which skips the checks. As for commands, /compact and /context work, but /clear doesn't; you start a new session from the sidebar instead. /model and /effort don't open a picker, so you type them with a value, as in /model sonnet.

If what you want is to drive your own machine, with its files and network, from somewhere else, then the tool you want is Remote Control, not the cloud. Our Remote Control guide compares the two in detail.

3. Getting started from the browser or the terminal

There are five ways in: claude.ai/code in a browser, the Code tab in the Claude mobile app, a session in the desktop app started with Cloud selected instead of Local, claude --cloud in the terminal, and routines (runs triggered on a schedule or by an event). The official guidance is to do your first one from the browser.

From the browser (your first session)

1. Open claude.ai/code and sign in
2. Connect GitHub: follow the prompts and approve on GitHub's authorization screen. This lets you work with public repositories
3. To use private repositories, install the Claude GitHub App: you can only work in private repositories where the app is installed. On a GitHub organization, an organization owner may need to approve it
4. A Default environment is created: on Pro and Max it's created automatically. On Team and Enterprise you'll see a creation form; leave the defaults and click "Create & finish"
5. Pick a repository, a branch and a permission mode, then write and send your task

For writing tasks, the official documentation recommends naming the file or function rather than just saying "fix tests", pasting error output if you have it, and describing the expected behavior rather than just the symptom. Each task gets its own session and its own branch, so you can send the next one without waiting for the previous one to finish.

From the terminal

If you already use the GitHub CLI (gh) locally, you can also type /web-setup inside Claude Code. It sends your gh token to Anthropic, where it's stored encrypted in your claude.ai account, and any repository that token can reach becomes usable without installing the app. On Team and Enterprise, the command doesn't appear until an organization Owner turns on "Quick web setup".

Once you're connected, you can send tasks straight from the terminal.

# Start a new cloud session for the repository you're in
claude --cloud "Fix the authentication bug in src/auth/login.ts"

# Send a follow-up instruction to a running cloud session
claude -p "Add tests too" --cloud <session-id>

# Pull a cloud session into your local terminal
claude --teleport

A session started with --cloud clones your current branch on GitHub, not your local working tree. Commits that exist only on your machine won't be there, so push first. You also can't move a local session to the cloud from the CLI (only the other direction, with --teleport). The desktop app does have a "Continue in" menu that sends a local session to the cloud.

To use this from the terminal, Claude Code must be signed in with a claude.ai account. With API key authentication, or when it's configured for a third-party provider such as Amazon Bedrock or Google Cloud's Agent Platform, neither --cloud nor --teleport works.

4. Environments, network access and secrets

Every cloud session runs in a "cloud environment". An environment is a saved configuration that bundles the network access level, environment variables and a setup script. The Default environment you start with has Trusted network access (see below) and no environment variables or script.

What's inside the VM

In Anthropic-managed environments, each session gets a fresh virtual machine running Ubuntu 24.04 (x86_64). Even if you're on a Mac or Windows, it's Linux inside, so dependencies with native binaries use the x86_64 Linux builds. Python, Node.js (20/21/22), Ruby, PHP 8.3, Java 21, Go, Rust, C/C++, Docker, PostgreSQL 16, Redis 7.0 and more come preinstalled. The approximate resources are 4 vCPUs, 16GB of RAM and 30GB of disk, and the official documentation notes that these may change. Work that goes well beyond the memory, such as a large build, can be killed.

Tools that aren't installed go in the environment's setup script (Bash, run as root). The script runs once in the first session, after which the filesystem state is saved and reused, so later sessions start faster. The requirements are that the script should finish in about five minutes and must not fail (a non-zero exit code means the session won't start). The saved state is rebuilt roughly every seven days.

The 4 network access levels

Note that None doesn't cut everything off. Traffic to GitHub goes through a dedicated proxy, any MCP connectors you enable go through Anthropic's servers, and Claude Code itself talks to the Anthropic API. The official documentation states it plainly: "When running with network access disabled, Claude Code can still communicate with the Anthropic API, which may allow data to exit the VM."

Where to put secrets

• Your GitHub credentials never enter the VM: a proxy swaps in scoped credentials on the session's behalf. git push only works to that session's working branch
• Environment variables can be read by everyone who uses the environment: the official settings dialog carries a note warning against putting secrets there
• Put API keys in "API credentials" (Pro and Max only): Anthropic's proxy attaches the key only to requests going to the hosts you specify. The key is visible neither to Claude nor to the commands it runs. It isn't available on Team and Enterprise yet

5. Pricing: your regular plan usage plus a limited-time credit

The pricing model is simple. The official documentation says cloud sessions share rate limits with all other Claude and Claude Code usage within your account, and states that there is no separate compute charge for the cloud VM. Run tasks in parallel and your limits go down proportionately faster.

At 10:57 a.m. JST the same day, @ClaudeDevs posted a follow-up that opened with an apology for the confusion. The gist: cloud sessions run on your Pro or Max plan like the rest of Claude Code, and this credit is an optional one-time promotion that cloud sessions spend first before falling back to your normal plan usage.

What local Claude Code shows when you hit your limit, and what to do about it, is covered in our usage limit article.

The full text of the offer terms (Claude Promotional Credit Offer Terms) contains several provisions worth knowing before you claim.

• The credit has no cash value and can't be transferred
• If you cancel your subscription, downgrade to the free plan or your account is suspended, the credit is forfeited and not restored, unless otherwise stated
• It's granted in your billing account's currency; if that differs from the advertised currency, Anthropic converts it at a rate it chooses and may round it to a convenient amount
• Anthropic can change or end the offer without notice, and can revoke credits or remove your eligibility if it determines there has been fraud, abuse or a violation of its terms. Anthropic has the final say on eligibility and amounts
• By taking part, you agree that Anthropic may turn on "Usage Credit" for your account if it isn't already enabled

That last point is easy to miss, so after claiming, it's worth opening the billing settings on claude.ai once to check that everything is set the way you intend.

6. How to claim the credit, and a note on accounts

The steps are in an @ClaudeDevs post.

1. In a browser, open the claim page (claude.ai/code/claim-credit). In the terminal, type /claim-credit in Claude Code
2. Connect GitHub: the post says "You'll need GitHub connected to start a session." If you haven't connected it yet, follow the steps in "3. Getting started"
3. Start a cloud session: the credit is spent first automatically

Note that as of September 24, 2026, /claim-credit isn't yet listed in the commands documentation or the CHANGELOG. The only source is the @ClaudeDevs post. If the command doesn't show up when you type it, claiming from the browser page is the reliable route.

7. Where cloud sessions fit, and where they don't

The official documentation draws a clear line between the jobs cloud sessions are good at and the ones that should stay on your machine.

On repositories outside GitHub: for a repository with no remote, or a github.com repository without the Claude GitHub App installed, claude --cloud starts the session by uploading your local repository as a bundle. You can do the same with GitLab or Bitbucket by setting CCR_FORCE_BUNDLE=1. The bundle has to be under 100MB, untracked files aren't included, and you can't push the results back to that GitLab or Bitbucket repository.

If you need your local environment but still want to run things in parallel, agent view, which runs multiple sessions side by side on your own machine, is the better fit.

8. Things to know before you start

Public sharing means "anyone logged into claude.ai"

On Pro and Max, a session's visibility is either Private or Public. Public makes it visible to any user logged into claude.ai, and repository access verification is not enabled by default. The official documentation warns: "Sessions may contain code and credentials from private GitHub repositories." The settings to require recipients to have repository access, or to hide your name, are under Settings > Claude Code > Sharing settings on claude.ai.

Auto-fix replies as your GitHub account

In repositories with the Claude GitHub App installed, you can use Auto-fix, which watches a PR and responds automatically to CI failures and review comments (from the terminal, it's /autofix-pr). Those replies are posted under your GitHub account (labeled as written by Claude Code). In repositories that use Atlantis, Terraform Cloud or GitHub Actions triggered on issue_comment, a comment can trigger infrastructure changes, so the official documentation suggests turning Auto-fix off for those repositories.

Secret files when uploading from your machine

When your local repository is uploaded as a bundle on macOS, Linux and WSL, uncommitted changes to files named like .env, *.tfvars, id_rsa or *.pem are left out. Put the other way round, committed versions are uploaded, and the three environments above are the only ones the official documentation names for this exclusion. It doesn't say whether the same exclusion applies on native Windows, so check what you're sending before you send it.

Other points

• Deletion can't be undone: deleting from the sidebar or the session menu permanently removes the session and its data. If you just want to tidy up, archiving is enough
• If your organization uses an IP allowlist, everything fails: cloud sessions call the API from Anthropic's infrastructure, so you'll need to ask support for an exemption
• User-level SessionStart hooks don't come along: hooks in ~/.claude/settings.json stay on your machine. Anything you want to run in the cloud too belongs in the repository's settings

9. Cloud sessions vs Projects: who does what

Claude Code's Projects, rebuilt on September 17, also run in the cloud. Under the hood, each "thread" in a project is a cloud session. The difference is who hands out the tasks.

The fact that the credit only covers cloud sessions feeds directly into how you split the work. If you're trying things out on the credit, start with cloud sessions, sending tasks yourself one at a time, and move to Projects once the work settles into something longer-running. That's the natural order. Our Projects article goes into its rollout conditions and how fast it uses tokens.

Summary

Cloud sessions are the feature for running Claude Code detached from your own computer. Hand over a GitHub repository and write a task, and Claude works in a cloud VM, pushes a branch and waits. It keeps going when you close your laptop, you can reply from your phone, and there's no separate charge for the VM.

In exchange, your local config and tools don't come along, GitHub is assumed, and you can't approve each step. The most straightforward split is to keep work that needs your local environment on your machine or on Remote Control, and to send well-defined one-off tasks and anything you want to run in parallel to the cloud.

If you're an existing Pro or Max subscriber, claim by October 7 and the credit is spent first until November 5, 16:59 JST. It's a window where you can experiment without worrying about your limits, so start by sending one small task and see how the diff looks and how long it takes to come back.

FAQ

Q. Can I use it on the free plan?

A. No. The official documentation lists Pro, Max and Team, plus Enterprise users with premium seats or Chat + Claude Code seats. This credit is also only for existing Pro and Max subscribers.

Q. Do cloud sessions cost extra?

A. There's no separate compute charge for the VM. They draw on your plan's usage limits, the same limits as local Claude Code and chat. If you've claimed the credit being handed out from September 24, the credit is spent first, and once it runs out or expires you go back to your normal limits.

Q. Can I use the credit on Projects or Routines?

A. No. The claim screen says Projects and Routines are excluded. Routine runs also execute as cloud sessions, but they're left out of what the credit covers.

Q. Can I try it without using GitHub?

A. You need GitHub connected to start a session. That said, claude --cloud can also start a session by uploading a local repository that has no GitHub remote (under 100MB, and results can't be pushed back to GitLab or similar). If you don't want to connect GitHub, the alternative the official documentation points to is Remote Control, which runs on your own machine and lets you follow along from a browser or phone.

Q. Can I move work I've started locally to the cloud?

A. Not from the CLI. --teleport only pulls a cloud session down to your machine. --cloud creates a new session and clones your current branch on GitHub. The desktop app has a "Continue in" menu that sends a local session to the cloud.

Q. What happens if Claude asks me a question and I leave it hanging?

A. Until the environment expires, you can come back, answer and carry on from there. Even after it expires and the VM is reclaimed, reopening the session from claude.ai/code provisions a fresh VM with your conversation history. Any subagents or shell commands that were running in the background at the time won't be restored, though.