Boxes.dev vs Codex
Both run Codex in the cloud and let you start work from Slack, Linear, or your phone: Codex Cloud on a machine for each task, started from an environment Codex prepares from your repositories, and boxes.dev on full, isolated copies of your development environment, with Claude Code alongside.
The short answer
Codex Cloud runs each task on its own machine, started from a reusable environment Codex sets up from your GitHub repositories, within your ChatGPT plan's usage allowance. Boxes.dev runs Codex, alongside Claude Code, on a full copy of the cloud machine where your app already runs. Choose Codex Cloud for a simple way to run Codex remotely or automatic reviews on pull requests. Choose Boxes.dev when you want the agent to test its work in your full running app, like an engineer would, to work in the same machine yourself, or to run Codex and Claude Code side by side.
Choose Boxes.dev if
- You want the agent to test its changes in your running app, with its real databases, services, and data, the way an engineer would.
- You want to open a terminal on the agent's machine, edit files, and use the running app in a browser you share with the agent.
- You want to come back to the same machine days later, with its dev servers and data where you left them.
- You want Codex and Claude Code in one app, each signed in with your own plan.
- You need to share a running app with a teammate, or give an OAuth provider or webhook one stable public URL.
Choose Codex if
- You want a simple way to send a task from your terminal to the cloud with one command.
- You want automatic Codex reviews on every new GitHub pull request.
- You want a security agent that finds and confirms vulnerabilities in your repositories.
At a glance
As of October 2, 2026. This page compares boxes.dev with Codex Cloud, where OpenAI runs each Codex task on its own virtual machine, using the reusable cloud environments OpenAI introduced on September 29.
| Boxes.dev | Codex | |
|---|---|---|
| Agents | Codex and Claude Code, each with your own plan or an API key, in a graphical view or the agent's own terminal UI | Codex, with your ChatGPT plan |
| Where cloud work runs | A devbox for each task: a full, isolated copy of your Template box that can hold several repositories and threads | A virtual machine for each cloud task, run by OpenAI and started from a published cloud environment |
| Cloud environment | A snapshot of your Template box, a cloud machine set up with your app's tools, databases, services, and data, kept current by an optional maintenance script | The files of your GitHub repositories with the dependencies and tools Codex installed and tested with you, saved when you publish and updated when you republish |
| Between tasks | A devbox sleeps, keeping files and memory, so dev servers, databases, and terminals resume when it wakes, and it stays until you delete it | A task keeps its files, uncommitted changes, and installed tools, which can be recovered for up to seven days after you last use it |
| Working in the cloud machine | Terminal tabs, a file editor, Git history and diff review, a shared browser, a process monitor, and SSH | Changed files and test results to review, follow-up requests, and a commit or pull request when you're ready |
| Reaching the running app | A browser on the devbox that you and the agent share, with comments on the page, ports forwarded to the same port on your laptop, stable .localhost addresses, public links, and Project URLs | For apps running on your computer, the desktop app's browser previews pages, with comments on the page; OpenAI lists browser use in cloud environments as not yet supported |
| Apps | macOS desktop app, iOS and Android apps with a terminal, the dvb CLI, SSH, and VS Code or JetBrains through dvb ide | The ChatGPT desktop app, Codex on the web, the ChatGPT app for iOS and Android, the Codex CLI, and IDE extensions |
| Starting work | The apps, dvb codex or dvb claude from your terminal, @boxes.dev in Slack, assigned Linear issues, and automations on a schedule or from an API call | The apps, codex cloud from your terminal, @ChatGPT in Slack or Microsoft Teams where an admin turns it on, assigned Linear issues, and @codex on GitHub pull requests |
| Pull requests | Comments on changed lines go to the agent, and pull requests can fix failed checks and bot review feedback on their own | @codex review and automatic reviews flag serious issues, and other @codex comments start a cloud chat that can push a fix |
| Shared setup | Environment files with project, team, and personal values; startup and maintenance scripts; MCP servers and agent plugins on every devbox; AWS, Tailscale, and a static egress IP; Team Templates | Cloud environments with environment variables, secrets that a proxy adds to requests for allowed services, values each person supplies from a personal vault, internet access settings, Tailscale, and OIDC cloud credentials on Enterprise |
| Working with your team | Read-only thread links, team-wide thread reading, and project browsing for teammates; @boxes.dev in Slack, where teammates you allow can start work and send follow-ups that run with your access; commits linked to threads; and Team Templates | Read-only snapshots of local threads, shared by link from the macOS app; cloud environments shared with your workspace, where each person starts their own tasks; @ChatGPT in Slack, where each person's tasks run with their own account |
| Agents working together | Agents start agents on new or existing devboxes, message each other, read earlier threads, and share files in a project folder | Subagents work in parallel within a chat and report back in one response |
| Pricing | Free trial with 20 box-hours; Starter $19 and Pro $99 per user each month, plus your ChatGPT or Claude plan or an API key | Included in ChatGPT Plus, Pro, and Business, from $20 a month; cloud tasks count toward the plan's usage allowance |
What the cloud machine starts from
Both start each task from a saved setup: Codex Cloud from an environment that Codex prepares from your GitHub repositories, and boxes.dev from a copy of a machine where your app already runs.
To create a Codex Cloud environment, you choose GitHub repositories, and Codex inspects them, installs their dependencies and tools, and tests the setup with you, asking for any access or values it needs. It records an install script and a start skill, which tells the agent how to start your services and check that they're ready. When you publish the environment, OpenAI saves its prepared files, and each new task runs on its own virtual machine started from them. To change the setup, you edit the environment with Codex and publish it again. Cloud tasks count toward your ChatGPT plan's usage allowance, with no separate charge for the machine.
Boxes.dev first sets up your full development environment on a cloud machine called the Template box, with the tools, databases, services, and data your app needs, so an agent can run the app and test its changes end to end. A setup agent can port the environment from your laptop, including the files, local data, agent settings, and skills you choose to upload, or build it from a GitHub repository. Boxes.dev then saves the Template box as a snapshot, and every task runs on its own devbox, a full, isolated copy of that machine. A maintenance script can keep the Template box current as your code changes, running on a schedule or when files such as a lockfile change, and each successful run saves a new snapshot.
You'd never let an engineer ship code they hadn't run, and a full environment holds the agent to the same standard. On a boxes.dev devbox, the agent runs your app and tests its changes in the browser as it builds, and it reproduces bugs in the real app to find the root cause instead of guessing from the code, so it writes better code and makes fewer mistakes. It can also take screenshots and record video for you to review, so you can hand it most repetitive manual testing.
What's kept between tasks
A boxes.dev devbox keeps its running programs through sleep and stays until you delete it, and a Codex Cloud task keeps its files for up to seven days after you last use it.
Each Codex Cloud task keeps its own files, including uncommitted changes and installed tools, so you can reopen it from the web, the ChatGPT mobile app, or the desktop app and continue. OpenAI's docs say a task's saved state can be recovered for up to seven days after you last start a turn or resume it. New tasks start from the published environment, and OpenAI refreshes its repositories in the background.
A boxes.dev devbox sleeps when nobody is using it, and an agent that is still working keeps it awake. Sleep keeps the machine's files and memory, so dev servers, databases, and terminal sessions resume where they left off when it wakes. You can come back the next day to the same files, branches, and data.
Working in the machine
In Codex Cloud, Codex runs the commands on the task's machine, and on boxes.dev, you can also work in the devbox yourself.
In a cloud task, you describe the goal, and Codex edits files, runs commands, and checks its work. You can then review the changed files and test results, ask for follow-up changes, and commit or open a pull request. The desktop app's integrated terminal works in projects on your computer.
On boxes.dev, you can open a terminal on the devbox, edit its files, and run
your app alongside the agent, with the tools listed under
What comes with every boxes.dev devbox,
over SSH, or in VS Code or a JetBrains IDE with dvb ide.
Reaching the running app
On boxes.dev, you and the agent can both use the app running on a devbox, and the ChatGPT desktop app previews apps running on your computer.
The ChatGPT desktop app's built-in browser shows you and Codex web apps running on your computer, and you can comment on the page for Codex to address. OpenAI's cloud environment docs list browser and computer use as not yet supported in Codex Cloud.
Each boxes.dev devbox runs a browser that you and the agent share, so you can
watch the agent test the app, click through it yourself, or comment on page
elements to show the agent what to change. Ports forwards the app to the same
localhost port on your laptop, a public link shares a devbox's running app
with anyone, and a Project URL gives your project one stable public address
that you can point at any devbox, which suits OAuth callbacks and webhooks.
Working with your team
Both bring agents into Slack, and boxes.dev also lets teammates you allow send follow-ups to your conversation and start work in your project, with your access.
In ChatGPT workspaces where an admin turns on Cloud delegation, mentioning
@ChatGPT in Slack or Microsoft Teams can hand repository work to a Codex
Cloud task, using an environment shared with the workspace. The task runs with
your account, and ChatGPT posts the result in the thread. Your later requests
in that thread can continue the task, and a request from someone else may
start a separate task with their account. You can also share a cloud
environment with your workspace so colleagues can start their own tasks from
the same setup, without access to each other's tasks. In the ChatGPT desktop
app for macOS, you can share a read-only snapshot of a local Codex thread by
link.
Each boxes.dev devbox belongs to one person and works with their access, such as their GitHub login and cloud credentials. Anyone who can prompt its agent could act with that access, so in the app, teammates get read-only access to your work. You can share a thread as a link, and a team admin can let teammates read every thread that isn't marked private and browse each other's projects in a Team workbench. Teammates can't send prompts, open your terminal, or control your devbox.
In Slack, mentioning @boxes.dev in a thread starts Codex or Claude Code on a
new devbox, and later mentions in that thread send follow-ups to the same
conversation. By default, only you can start work in your project from Slack
or follow up on it. If you allow teammates, they can also start work there
with @boxes.dev @YourName and send follow-ups, and their requests run with
your access and usage. Boxes.dev sends you a DM with links to each task a
teammate starts, and commits can link to the thread that produced them.
What comes with every boxes.dev devbox
The boxes.dev desktop app opens the same set of tools for whichever devbox you select:
- Terminal, with tabs that keep running after you close the app.
- Files, a file browser and editor with search across the workspace.
- Git/Review, for branch history and diffs, where comments on changed lines go to the agent together as one message.
- Browser, the browser on the devbox that you and the agent share.
- Ports, for the forwarding,
.localhostaddresses, and public links described above. - Activity, which shows the processes using the machine's CPU and memory.
AWS, Tailscale, and the MCP servers and Codex plugins you add for
the project reach every devbox, and agents can leave each other notes in a
shared /project folder. You can branch a conversation onto a new devbox that
starts with the current files, including uncommitted work.
Where Codex shines
- One command from your terminal.
codex cloud execsubmits a task to a cloud environment, andcodex applybrings a cloud chat's latest diff into your local repository. - Automatic pull request reviews. Codex can review every new GitHub pull
request, flag only high-priority issues, and follow review rules from
AGENTS.md. - Credentials kept out of cloud tasks. Programs in a task see a placeholder for each network secret, and a proxy adds the real value only to requests for the services you allow.
- Codex Security. A security agent finds, confirms, and helps fix vulnerabilities, from the desktop app, the terminal, an SDK, or connected GitHub repositories.
Pricing
As of October 2026.
| Boxes.dev | Codex | |
|---|---|---|
| Price | Trial free; Starter $19 and Pro $99 per user each month | ChatGPT Plus $20 a month; Pro from $100; Business $25 per user each month |
| Model usage | Your ChatGPT or Claude plan, or an API key | Included in the plan's usage allowance, shared by local and cloud work |
| Cloud time included | 20 box-hours on the Trial, 40 per user on Starter, 250 on Pro | No separate charge; cloud tasks use the plan's allowance and may use more of it than local messages |
| Cloud machines awake at once | 2 on the Trial, 4 per user on Starter, 10 on Pro | No limit is published |
| Extra cloud time | $0.75 per box-hour on Starter, $0.60 on Pro | ChatGPT credits extend the plan's usage |
A box-hour is one default devbox awake for an hour, and sleeping devboxes don't use any. On Starter, 40 box-hours works out to about two hours of devbox time each working day, and Pro's 250 to about twelve, spread across as many as ten devboxes at once. Your ChatGPT plan covers Codex's usage on boxes.dev, and the boxes.dev plan covers the machines. See Plans, seats, and box-hours.
OpenAI's pricing page says that weekly limits may also apply to the plan's usage allowance, and that Business costs $20 per user each month when billed annually. See Codex pricing.
Using both
We recommend using both. Codex is a great agent, and boxes.dev runs the real
Codex CLI on every devbox with your ChatGPT subscription, so your AGENTS.md,
skills, and settings work the same way in both places. You can keep Codex's
automatic reviews on your pull requests and Codex Cloud for quick tasks from
your phone, and use boxes.dev for tasks that need your full environment, a
terminal, or the running app.
Frequently asked questions
Is boxes.dev affiliated with OpenAI?
No. Boxes.dev is an independent product and isn't affiliated with or endorsed by OpenAI.
Can I use my ChatGPT plan with boxes.dev?
Yes. Sign in with ChatGPT to connect Codex, or use an API key. Codex's usage on boxes.dev counts toward your ChatGPT plan, and boxes.dev charges only for machine time. See Connect Codex and Claude.
How does a Codex Cloud environment compare with a Template box?
Both save a prepared setup that each new task starts from. A Codex Cloud environment holds the GitHub repositories, dependencies, and tools that Codex set up and tested with you, and each task keeps its files for up to seven days after you last use it. A boxes.dev Template box is a full cloud machine where your app already runs, a maintenance script can keep it current, and each devbox copied from it stays until you delete it, with its running programs resuming when it wakes from sleep.
Do AGENTS.md, skills, and MCP servers work?
Yes. Codex on a devbox reads your repository's AGENTS.md, and project setup
can copy agent settings, skills, and MCP definitions from your computer.
Plugins that Codex syncs to your account work on devboxes too.
Sources
- Codex: Codex Cloud, cloud environments, Codex Cloud (Legacy), Codex CLI commands, GitHub code review, ChatGPT in Slack, Linear, integrated terminal, browser, sharing threads, subagents, Codex Security, and pricing.
- Boxes.dev: How boxes.dev works, Update your Template box and snapshots, Startup, teardown, and maintenance, Sleep, wake, and recover a devbox, Review a page with your agent, Preview your app, Project URLs, Shared project storage, Use the agent's terminal view, Install agent plugins, Organize and share threads, Team access and roles, Commit, push, and open a pull request, Install and use the CLI, Work from Slack, Work from Linear, Use boxes.dev on your phone, and Plans, seats, and box-hours.
More comparisons
Product names are trademarks of their respective owners.
