If the PC running Codex goes to sleep, you can't continue work that you're doing through Remote on that PC. Work you sent to Codex Cloud can continue even while your local PC is asleep. What matters is not whether you have the app open on your phone, but where the files are read and the commands are run.
Check where work runs before you close your PC
Give instructions and approvals from your phone
↓
A Mac or Windows PC that is awake
↓
Files and development environment on that PC
If the connected PC sleeps, access stops
Make a request from the web, phone, or PC
↓
Choose a published development environment
↓
Cloud workspace and tools
Can continue even when your local PC is asleep
Table of contents
- 1. Differences between Local, Worktree, Remote, and Cloud
- 2. Closing the window, sleep, and quitting the app are different
- 3. Steps to continue your local PC work from your phone
- 4. Codex Cloud when you want your PC to rest
- 5. Can local work move to Cloud as-is?
- 6. ChatGPT Work and dot compared
- 7. Pricing, usage limits, and how to choose
- 8. Checklist when your phone can't connect
- 9. What to check before stepping away and after returning
- FAQ
1. Differences between Local, Worktree, Remote, and Cloud
Local and Worktree both run on your PC. Worktree separates Git working areas; it is not a feature for moving execution to the cloud. Remote is a route for reaching the execution target from another device, and Cloud is a method that selects a cloud development environment as the execution target.
| Mode | Where files and commands run | When the PC is put to sleep |
|---|---|---|
| Local | The project folder on the PC you selected | Don't assume processing continues on that PC |
| Worktree | A separate Git working area on the same PC | Same as Local Separating working areas does not change power conditions |
| Remote | The connected Mac or Windows PC, and the environment you use from that PC | The connected PC must be awake, online, and running the app |
| Codex Cloud | A cloud working area created from a published environment | Work can continue even when your local PC is asleep |
| SSH remote projects | The file system and shell of the SSH host | From your phone, the connection goes through the host in the desktop app Running only the server does not guarantee the connection |
Sources: Codex environments and What Remote uses from the connected host. That a PC is the host does not mean that model computation and communication are completely contained on the device.
For example, if you connect from your phone to an always-on PC at home, the power state of the laptop you carry is not a condition. On the other hand, if the laptop itself is the Remote target, putting it to sleep makes it inaccessible. Don't judge by the phrase "my PC" alone; check the host name you selected and where the work is running.
2. Closing the window, sleep, and quitting the app are different
Whether work continues after you "close" something depends on what you close. Distinguish the phone screen, the laptop lid, and the app on the connected PC.
You have left the screen for giving instructions and checking. This is separate from whether the execution PC is available. You can return to the same task later to check the results.
Remote access stops. A setting that prevents sleep is not a guarantee that a sleeping PC will wake up automatically.
The official documentation lists this as a condition that stops Remote. Whether closing the window also quits the app depends on your OS and the app's state.
OpenAI's Remote guide explains that if the connected PC sleeps, loses its network connection, or the app is closed, Remote access stops until the host is available again. The official documentation does not explain how far a running command gets after disconnection, or what resumes automatically on reconnection. After resuming, check the task's final record, the change diff, and the test results separately.
When the Mac lid is closed
The official documentation says that with a MacBook, keeping the lid open and connected to power keeps Remote available. If you close the lid, it also states the condition of connecting an external display, and says that explicitly choosing Sleep stops Remote. "The lid is closed" and "the Mac is asleep" are different states.
When using Windows screen control
For work that uses Computer Use on Windows, the official condition is to keep the session unlocked and available. Screen control runs in the foreground, so the assumption is that you reserve that PC for this work. This is a condition for Windows screen control. It does not mean that every command execution always stops when the screen is locked.
Source: Remote: conditions for the PC you connect to. For work started from the CLI or an IDE extension, too, don't confuse closing the terminal, the OS going to sleep, and quitting the development tool.
3. Steps to continue your local PC work from your phone
If you want to use an existing local repository, tools you've already installed, and your browser sign-in state, Remote is the option to consider. Use your phone to give instructions, approve actions, and check results; the environment is provided by the connected PC. The steps in the official guide are as follows.
- Confirm the same account and workspace. Make sure the latest ChatGPT desktop app is running on your Mac or Windows PC, and that the PC is awake and online. In organizations, administrators may need to allow Remote Control.
- Open the connection settings on the PC. Go to Settings → Connections → Control this Mac or PC, select Set up or Add, and complete the access permission and any required identity verification.
- Scan the QR code with your phone. Complete the connection in the ChatGPT app. MFA, SSO, or passkey checks may be required. Connections are set up for each pair of phone and host.
- Open Codex on your phone. On iOS, open Codex; on screens that still use the old label, open Remote. Including Android, what appears and what you can use depends on rollout status. Select the target host and the existing chat.
- Check both the instruction you sent and the result. Review progress, diffs, terminal output, and test results. When an operation needs permission, check the approval request, and don't repeatedly tap the same request.
Sources: Prerequisites and Setup steps. This mobile connection setup starts from the desktop app. It is not a statement that the same setup can be done with only the CLI or an IDE extension.
Remote uses the connected host's credentials, plugins, browser environment, and permissions. Which PC you allow also affects the scope of your work. Don't assume that you must add screen control or browser extensions across the board just to connect. Decide the work you need and the permissions first.
4. Codex Cloud when you want your PC to rest
If you want to keep fixing code or investigating while your PC sleeps, prepare a repository, dependencies, and a test environment that can be used in the cloud. You can also start and continue work from your phone, but the condition is that you can access a published development environment first.
- On the web or in the desktop app, choose Work in → Cloud → Select environment → Create environment.
- Select the target GitHub repository and connect it if needed. Use Get started to proceed with preparing and checking dependencies and tools.
- Check the settings, required access, and test results, then choose Publish.
- Confirm Environment published, then make a request from Start a new task.
- From another device, open the same task to check the results or give additional instructions.
Publish here is an operation that saves the prepared state of the development environment for new tasks. It differs from publicly releasing a web app, but check the sharing scope of the environment in settings such as Who can use. Codex runs investigation, installs, and tests even during preparation, so don't treat it as "no processing happens because I haven't asked for development yet." Source: Create and publish an environment.
What is not automatically brought into the cloud
| What's on your machine | How Cloud handles it |
|---|---|
| Uncommitted code | Don't assume your local changes are transferred automatically Check the selected repository and its starting state |
| Dev servers and local databases | Locally running processes are not moved automatically Set them up on the Cloud side if needed |
| Browser sign-ins and VPN | Your PC's authentication state and VPN access are not shared automatically |
| Personal skills | Local personal skills are not synced Skills saved in the repository can be used |
| Changes to a published environment | Even if you republish the environment, existing tasks do not automatically change to the new starting state |
Sources: Execution location and resources and Reusing saved state. A new task starts independently from the published environment. Existing tasks carry over their own saved files and installed tools.
Limits on screen checks and how long state is kept
Current Cloud environments list Computer and browser use, as well as GitLab and self-hosted GitHub Enterprise Server, as unsupported. Sending work to Cloud doesn't always mean that operation checks in a real browser are included. Separately confirm whether tests such as Playwright run from the command line can run in the environment you need.
The saved VM state is described as recoverable for up to 7 days by default after the last turn started or the task was resumed. This does not mean it runs automatically for seven days. Protect important code with commits and the artifacts you need, and don't use the cloud's saved state as a substitute for source control. Sources: VM and saved state and Current limitations.
Codex Cloud (Legacy) on the settings screen is an older environment for Code Review, Linear, and GitHub integrations. Don't mix it with the new Cloud environments or their settings. Also avoid applying the legacy guide's explanations of container caches and setup secrets to the storage or authentication rules of the new environments. Source: Codex Cloud (Legacy).
5. Can local work move to Cloud as-is?
Handoff, described in the official Remote guide, is a feature that passes chat and Git state between your local PC and another connected host. The destination needs a saved project for the same Git repository. If you use a subdirectory, register the same location on both sides. Handing off while a task is running interrupts the current response.
The documentation clearly states that this Handoff does not support moving to a Codex Cloud environment. Moving to another Remote host and moving to Codex Cloud are different things. Even if the PC shuts down, an in-progress Codex chat does not automatically move to Cloud. Source: Handing off a chat between hosts.
- The repository, and the version or branch to start from
- Changes made so far, and changes not yet applied
- Tests that passed, tests that failed, and the conditions to reproduce them
- The remaining completion criteria, prohibited operations, and how to handle external publication
Sharing code and moving the whole conversation and execution state are different things. If you first ask Cloud to confirm the starting state, you reduce the chance of making changes on top of parts that are already fixed.
Keeping completion criteria in the same chat with How to use Codex /goal also does not change where work runs or the PC's power conditions. You need both a goal that keeps the work going and an execution environment that is available.
6. ChatGPT Work and dot compared
The ChatGPT Work approach of "coordinating in the cloud and using tools on an allowed PC" is different from Codex Remote. It applies only to new tasks that you start in the desktop app by selecting Cloud after enabling Local computer access with Work Cloud.
If the PC can't be used when a new turn starts, work may continue in a cloud container. However, the files and tools on that PC cannot be used. You cannot switch from local to Cloud in the middle of a turn.
Even if the dot itself continues later work in the cloud, child tasks on a PC that has become unavailable do not move to the cloud automatically. Work that requires that PC cannot proceed.
Sources: Distinguishing Work Cloud from Remote and End-user conditions for Work and dots. The latter is an Enterprise guide, and whether you can use it depends on your workspace. The role of dot is covered in How to use dot, its pricing, and how it compares with Codex.
7. Pricing, usage limits, and how to choose
Codex's pricing page explains that local messages and Cloud tasks share your plan's usage limits. Cloud tasks may use more of the limit than local messages, but for the same request there is no fixed multiplier. The usage varies with the model, the context, reasoning, tool use, retrieved information, caching, and the nature of the work.
The pricing table's explanation for Free and Go concerns the phased rollout of GPT-6 Luna in the desktop app. Don't conclude from that alone that you can use Codex Cloud or Remote for free. The Plus description includes Web, CLI, IDE, iOS, and cloud integration, and in organizations, administrator access settings are also a condition. The feature table lists mobile Remote Control with an API key as unsupported, and it is treated differently from SSH remote connections. The pricing description for API-key use does not include cloud features, and API pay-as-you-go charges cannot be converted directly into a number of tasks within a subscription. Source: Codex pricing and usage limits.
Buying additional credits after you reach the limit is a separate payment. Using credits you already hold and buying or automatically reloading new credits are also separate. Changing where work runs does not restore your usage limit. For Pro pricing tiers and reset options, see Pro 100, 200, and 500 compared.
| What you want to prioritize | Option | Check before choosing |
|---|---|---|
| Giving instructions from outside while keeping your current PC environment | Remote | Can it be used while the host stays awake? Can the browser, sign-ins, and tools be used there? |
| Letting the laptop sleep | Codex Cloud | Can the necessary code and test environment be prepared in the cloud? Will the limits on screen operation not get in the way? |
| Gathering your existing environment on an always-on PC | Remote to another host | Can you take on that PC's operation, access rights, and power and communication costs? |
| Using a development environment on an SSH server | SSH remote projects | Can you install and authenticate Codex on the server? Can the host that relays the phone connection also be used? |
For SSH connections, the official documentation requires that Codex be installed and authenticated on the remote side, and that codex be in the PATH of the login shell. It does not recommend exposing an unauthenticated app-server on a public network to keep a connection alive. Check the costs of dedicated servers, VPNs, and external services separately from the usage limits included in your Codex plan.
8. Checklist when your phone can't connect
Don't conclude that code or history on the PC has been deleted just because a connection failed. The following order lets you isolate the problem without first resetting the environment or deleting connection information.
- Check the target host. Make sure you are not connected to a different PC, account, or workspace.
- Check the PC's state. Check whether it went to sleep, lost its network connection, quit the app, or signed out.
- Check Remote Control after signing out. The official documentation explains that signing out turns Remote Control OFF, but device pairings are not deleted. Signing back in alone may not restore the earlier state.
- Separate waiting for approval from a connection failure. Check whether the target Codex or Remote task has an identity verification or approval request. Recheck that the account and workspace pair match.
- If it still doesn't work, record the state and the time. Note the OS, the app version, the connected host, the most recent sleep or sign-out, and any error shown. Leave confidential code, credentials, and personal information out of your report.
Source: Remote troubleshooting. When you try the restarts or reconfigurations that the official documentation suggests, save your in-progress changes and results first. Seeing the app after waking the PC and being able to send your next message are also separate checks.
A public GitHub report, user report #23470, describes a case where, after a Mac woke from sleep, Remote could not reconnect and returned 409 Conflict even though it was online. Since this is an individual report, don't treat deleting the internal settings mentioned in it as a recovery procedure you can follow as-is.
If you can read your history but sending fails and errors such as thread not found appear, also see a recovery procedure that separates saved history from execution state. Before you delete connection settings and start over, read the full text of the error you're getting.
9. What to check before stepping away and after returning
- Is the execution target This computer or Cloud?
- If Remote, is the connected PC awake?
- Are your changes and important artifacts saved?
- Have you specified the tests and stop conditions?
- Is the scope of permission for publishing, purchasing, and settings changes clear?
- Did you open the same host and the same task?
- What was the last operation completed, and what is still unfinished?
- Are the diffs and test results backed by evidence?
- Is anything waiting for approval, or limited by usage?
- Can you ask for only the unfinished part to be continued?
Use Remote if you want to work in the environment of an existing PC. Consider Codex Cloud if you want work to keep moving while your PC sleeps. However, Cloud has resources that cannot be moved to it and features it does not support, and Remote depends on the host's operating conditions. The key to choosing is not to treat "I can see it on my phone," "the conversation remains," "the work runs," and "verification is complete" as a single state.
Basis for this article: written on October 7, 2026, based on OpenAI's official documentation, including Remote connections and Cloud environments. PC sleep and connection drops were not reproduced on real devices.
Related articles
- Codex secret files, permissions, and network boundaries: what can be read in each execution location
- How to use Codex and Claude Code for read-only work: asking only for investigation and advice without changing code
- How to schedule ChatGPT tasks to run regularly: scheduling notifications and information organization that don't use a development environment
- Claude Code vs. Codex · How to use ChatGPT Work: choosing between products
FAQ
If I close the laptop I used to start Codex, can I keep going from my phone?
If that PC is the Remote target, it must be awake, online, and running the desktop app. If you close the lid and it goes to sleep, Remote access stops. For a Codex Cloud task, work can continue while your local PC is asleep.
Is Codex on my phone always a feature that connects to my own PC?
No. The official documentation also describes a route where you choose a published environment to start and continue a Cloud task. The Remote target host and the Cloud development environment are separate execution targets. Check the items you see against the app and the rollout status.
If my PC goes down, does Codex hand off to Cloud automatically?
Codex Handoff does not support handing off to a Cloud environment. Work Cloud target tasks have conditions for continuing in Cloud on a new turn, but the PC's local files and tools cannot be used, and it is not a feature for switching an in-progress turn partway through.
If I use Cloud and leave it alone, will it always finish?
Not needing to keep your PC awake and completion being guaranteed are different things. Authentication, approvals, missing tools, execution errors, or usage limits may require your action. Judge the result by the deliverables, the diffs, the checks that were run, and the items not yet verified, not just the final report.