How to Use Codex on Your Phone for Code Reviews and Approvals

Quick Answer

Yes, if the preview has reached your account. As of 2026-05-15, OpenAI describes Codex in the ChatGPT mobile app as a preview rolling out on iOS and Android across all plans, including Free and Go, in supported regions. When it is available, you can use your phone to start threads, steer active work, approve commands, review diffs, inspect test output, and check screenshots while the actual work runs on a connected host.

Safe setup checklist:

update the ChatGPT mobile app and the Codex App for macOS,
keep Codex running on an awake Mac host,
connect that host to the ChatGPT mobile app with the Codex mobile QR flow,
use the phone for approvals, review, and follow-up instructions,
keep risky edits, pushes, and merges behind explicit stop rules.

Use it to unblock an agent run away from your desk. Do not use it for deep code reading on a small screen, production access changes, or long debugging sessions that need a full desktop.

When To Use Codex Mobile

Use Codex mobile if you already run Codex on a Mac and need one of these from your phone:

approve a blocked command
check whether a background task is done
review a diff before you get back to your laptop
send a follow-up instruction after seeing test output
start a narrow task on a connected repo while traveling

Skip it if you want full local development from a phone. Codex mobile is a remote control surface, not a replacement for your editor.

May 2026: Mobile Preview And Host Rules

On 2026-05-14, OpenAI announced Codex in the ChatGPT mobile app as a mobile preview. OpenAI's Codex changelog also records that, on 2026-05-13, it added public Codex mobile documentation with setup steps, connected-host behavior, security requirements, and troubleshooting.

OpenAI's announcement, Codex changelog, and remote-connections docs set the practical scope:

the ChatGPT mobile app on iOS and Android can connect to a Codex App host
you can start, steer, approve, and review Codex work from the phone
the host must be a Mac running the Codex App and signed into the same account and workspace
the phone uses the host's projects, credentials, plugins, browser setup, permissions, and local tools
mobile access is a preview rollout, so some accounts or regions may not see it immediately
connecting a phone to the Codex App on Windows is not supported yet; OpenAI says Windows support is coming soon

Phone vs Host Execution

Your phone does not become the coding machine.

The phone sends prompts, approvals, and follow-up instructions. The connected host does the execution: reading files, running shell commands, loading project tools, accessing signed-in websites, and applying the same sandbox and approval rules that already exist on that host.

OpenAI's remote-connections docs are explicit about the boundary:

Surface	What it does	Why it matters
ChatGPT mobile app	starts threads, sends follow-ups, approves actions, reviews outputs	good for supervision and light steering
Mac host running Codex App	provides files, shell, credentials, plugins, browser access, and local tools	task execution happens here
Optional SSH host behind the Mac host	provides the remote filesystem and shell for the project	useful when the repo lives on another machine

Phone convenience does not remove host risk. If the host can reach a repo, browser session, or internal tool, Codex mobile can act inside that same boundary after you approve it.

Minimum Requirements

As of 2026-05-15, check availability before setup: OpenAI calls Codex mobile a preview rollout across all plans, including Free and Go, in supported regions. If Codex does not appear in the ChatGPT mobile app after an update, the feature may not have reached your account yet.

OpenAI's setup docs require all of the following:

Codex access in the ChatGPT account and workspace you want to use
the latest ChatGPT mobile app on iOS or Android
a Mac host that is awake, online, running the Codex App, and signed in to the same account and workspace
any required multi-factor authentication, SSO, or passkey flow completed during setup

Current limits:

mobile setup currently depends on the Codex App for macOS on the host
the setup flow is not available from Codex CLI or the IDE extension alone
Windows host support for phone connection is not available yet

If your team uses a ChatGPT workspace, an admin may also need to enable Remote Control access before the connection works.

Where Codex Mobile Helps

Good fits:

approving a dependency install after you have already reviewed the plan
checking whether tests passed on a background branch
asking Codex to revise one file after seeing failed output
reviewing a generated diff before your commute ends
starting a narrow triage or documentation task on a connected host

Weak fits:

broad refactors that require constant code navigation
frontend debugging where you need large-screen visual comparison
production changes with real blast radius
credential rotation, billing, or admin settings
any task where a mistaken approval would be expensive

Step 1: Set Up The Host Once

Start on the Mac where you already use Codex.

OpenAI's documented setup path is:

open Codex on the Mac host,
choose Set up Codex mobile in the sidebar,
scan the QR code with your phone,
finish the setup flow in the ChatGPT mobile app,
confirm the same ChatGPT account and workspace,
review host settings under Settings > Connections.

In the host settings, OpenAI says you can manage connected devices, choose whether to keep the Mac awake, and control features such as Computer Use or the Chrome extension.

If you plan to use this while away from your desk, keep the host plugged in and enable the setting that keeps the Mac awake. A sleeping host leaves the phone with nothing to control.

Step 2: Start With Review-First Tasks

Do not begin with an open-ended coding request.

Your first phone-supervised task should end in something you can inspect quickly on a small screen:

one diff
one test result
one review summary
one approval request
one short follow-up list

A good starter prompt:

Review the open task on this branch.

Rules:
- inspect the changed files first
- summarize the likely impact in 5 bullets or fewer
- run the smallest relevant checks
- stop and ask before installing dependencies, pushing, or changing more than 3 files

Step 3: Use Your Phone For Three Jobs First

The mobile app is strongest for supervision, not for full authoring.

1. Approvals

OpenAI says the phone can approve commands and other actions.

Use approvals from the phone when:

you already understand the task
the command is expected
the affected scope is small
you can verify the result from logs, tests, or a diff

Do not approve from the phone when the command changes credentials, production state, billing, or access control.

2. Follow-up steering

After a failed test or partial result, the phone is good for one clear correction.

Use short follow-up prompts like this:

The test failure on auth looks unrelated to the requested change.
Only continue if you can isolate the fix to the touched files.
Otherwise stop and summarize what needs desktop review.

3. Review

OpenAI's docs say mobile can review outputs, diffs, test results, terminal output, and screenshots. Use those views for queue-clearing tasks:

check whether the diff stayed inside scope
confirm that the failing command now passes
see whether a screenshot proves the UI change
decide whether the task needs desktop follow-up

If the review requires scrolling across too many files, stop and return to the desktop later.

Step 4: Write Better Mobile-Safe Stop Rules

Use a short rule block in tasks you expect to supervise from your phone:

Stop and ask before:
- installing new packages
- editing more than 3 files
- changing auth, billing, or deploy settings
- pushing, merging, or opening a PR
- using browser or computer control outside the named task

The mobile app inherits the host's tools and permissions. Before approving from your phone, confirm you understand which tools and credentials the command will touch.

Step 5: Pick The Right Host Model

OpenAI documents three practical host patterns.

Host pattern	Best for	Main tradeoff
Your everyday Mac laptop	quick access to the same projects, credentials, and tools you already use	remote access stops when the laptop sleeps, loses network, or closes Codex
A dedicated always-on Mac	longer-running tasks and higher availability	extra machine to maintain
A remote development environment through SSH	repos that already live on a devbox or remote machine	more setup and more security surface

If you only need occasional approvals while commuting, your normal Mac is enough. If you want to rely on phone access for longer background runs, the always-on host is a better choice.

Step 6: Use Mobile For Small, Copyable Tasks

Use prompts where the result fits on one screen.

PR review follow-up

Check the latest PR feedback on this branch.

Return:
1. comments grouped by file
2. smallest fix for each comment
3. checks to run after the patch

Rules:
- do not push
- do not broaden the scope
- stop if a comment implies redesign instead of a patch

Background task check-in

Summarize the current state of the running task.

Return:
- what is done
- what is blocked
- latest test result
- whether you need an approval from me now
- the next smallest step

Safe documentation update

Draft the requested docs update in the current repo.

Rules:
- stay inside docs files only
- keep the change under 150 lines if possible
- stop before opening a PR
- include a short verification note in the draft

Failures To Plan For

The host is technically connected but practically unavailable

If the Mac sleeps, closes Codex, or loses network access, the phone session disconnects. OpenAI calls this out directly in the troubleshooting docs.

The task outgrows the screen

A phone is good for decisions, not for deep repository reading. Once you need to compare many files or reason through a long trace, the task belongs back on the desktop.

The host has more reach than you remembered

Because mobile inherits the host's credentials, browser sessions, plugins, and local tools, a casual approval can have more effect than it looks like from the phone. Treat the host as the trust boundary.

Codex Mobile Or Claude Remote Control?

As of 2026-05-15, Codex mobile controls a Codex App host from ChatGPT mobile, while Claude Remote Control attaches to a local Claude Code session from Claude mobile or claude.ai/code.

Question	Codex mobile	Claude Remote Control
Phone entry point	ChatGPT mobile app	Claude mobile app or `claude.ai/code`
Local runtime	Codex App host, currently macOS for phone setup	Claude Code process started from CLI or VS Code
Setup trigger	`Set up Codex mobile` in the Codex App sidebar, then QR scan	`claude remote-control`, `claude --remote-control`, `/remote-control`, or VS Code `/remote-control`
Best fit	Teams already using Codex threads, worktrees, app review tools, ChatGPT accounts, and Codex host settings	Developers already coding in Claude Code terminal or VS Code who want to continue the same local session from another device
Main caveat	Preview rollout; the Mac host must stay awake, online, and signed into the same account and workspace	Research preview; the local process must keep running, Team/Enterprise admins can disable it, and plan eligibility should be checked in the current Claude docs

Claude's Remote Control docs say the web and mobile surfaces are a window into a local Claude Code session, while OpenAI's Codex docs make the ChatGPT mobile app a controller for a Codex App host or a remote environment connected through that host. In both designs, the work stays off the phone; neither turns the phone into the development environment.

Pick mobile when the bottleneck is approvals and check-ins, not when the bottleneck is writing code on a small screen.

FAQ

Can you code entirely from your phone with Codex?

No. The phone is a remote control and review surface. The actual coding work runs on the connected host.

Does Codex mobile work with Codex CLI alone?

No. As of 2026-05-15, OpenAI's remote-connections docs say mobile setup currently requires the Codex App for macOS on the host. The setup flow is not available from the CLI or IDE extension alone.

Does Codex mobile work with a Windows host?

Not yet for phone-to-host setup. OpenAI's 2026-05-14 announcement says support for connecting a phone to the Codex App on Windows is coming soon.

Is Codex mobile available on every ChatGPT account?

Not necessarily on the same day. OpenAI describes Codex mobile as a preview rolling out on iOS and Android across all plans, including Free and Go, in supported regions. Update both apps first, then check whether Codex appears in the ChatGPT mobile app.

Do you need the same account on both devices?

Yes. OpenAI says the host and phone must use the same ChatGPT account and workspace.

What is the safest first mobile task?

A review-first task with a clear stop line, such as checking one diff, approving one expected command, or requesting a short status update from a background run.

When should you skip Codex mobile?

Skip it when you need long code reading, sensitive admin changes, broad refactors, or any task where a mistaken approval has large consequences.

Verification Note

Verified on 2026-05-15.

Checked items: Codex mobile documentation added on 2026-05-13; Codex mobile preview announced on 2026-05-14; preview rollout on iOS and Android across all plans, including Free and Go, in supported regions; Windows phone-to-Codex-app support marked as coming soon; mobile control available through the ChatGPT mobile app; supported actions including starting threads, steering work, approving commands, and reviewing outputs; requirement for the same ChatGPT account and workspace; requirement for a Mac host running the Codex App; setup flow via QR code from the host; optional SSH-backed remote environment behind the host; host inheritance of files, credentials, plugins, browser setup, permissions, and local tools; troubleshooting notes for missing hosts, missing approval prompts, and host disconnects; and Claude Remote Control comparison points from the current Claude Code docs.

Official sources: OpenAI announcement, Codex changelog, Codex app, Remote connections, Codex product page, and Claude Remote Control docs.