Codex cloud environments are OpenAI's new reusable setups for coding tasks that keep running when your laptop sleeps. OpenAI announced them at DevDay on September 29, 2026. Each published environment holds repos, dependencies, tools, and access rules.

I'm Nitin from Debugged Pro. I work with software and IT agencies in India and the USA. This guide covers what changed, how security works, and a checklist you can use with clients this week.

OpenAI Codex Cloud Environments product art from DevDay 2026 — Codex Cloud Environments artwork from OpenAI DevDay 2026. Image credit: OpenAI

What Codex Cloud Environments Actually Do

Codex could already spin up remote cloud tasks. The DevDay update makes those setups persistent and shareable. You prepare once. New tasks start from that approved filesystem instead of a blank sandbox.

According to TechCrunch's DevDay coverage, developers can run Codex from a computer, phone, web, or the desktop app. The goal is faster task starts and a shared workspace with approved settings.

SiliconANGLE notes the real target is the prep work before an agent writes code. Dependencies are already installed when a task begins.

How you create and publish one

OpenAI's docs walk through a simple flow on web or desktop:

Choose Work in > Cloud, then create an environment.
Connect the GitHub repositories Codex should check out.
Let Codex inspect the code, draft the install script, and propose a start skill.
Supply missing access, review the setup report, then publish.
Start a new task from that published environment.

Saving stores configuration. Publishing captures the prepared filesystem for new tasks. Sharing controls who can use the setup.

Each task stays isolated

Every new task gets its own VM and workspace. Uncommitted changes stay with that task. You can reopen it later from a phone or browser. Dependency caches survive background repo refreshes. Existing tasks keep their own state when you republish the environment.

OpenAI DevDay screenshot showing Codex cloud environments and API updates — Codex and API updates shown at OpenAI DevDay. Image credit: OpenAI / TechCrunch

Why Codex Cloud Environments Matter to Agencies

Agencies lose hours every week recreating client stacks. New joiners install the wrong Node version. Contractors miss a private registry token. Codex cloud environments turn that setup into a published, approved baseline.

That matters for retainers and fixed-bid work. You can share one environment with a ChatGPT workspace so colleagues start their own tasks from the same stack. Access to the setup does not open someone else's task files. Enterprise admins hand out edit rights separately.

If you already advise clients on agent tooling, pair this with our OpenAI Dots agents agency checklist. Dots covers always-on personal agents. Codex cloud environments cover coding work that needs a real project toolchain.

USA agencies can standardize client sandboxes

Many USA shops juggle several GitHub orgs. Publish one environment per product line. Keep edit permission with tech leads. Let juniors and contractors run tasks without changing the shared install script.

India agencies can cut setup drag on US-facing work

Indian delivery teams often join US client repos across time zones. A published environment means a developer in Pune or Bengaluru can start a task from phone or web without waiting for a local machine rebuild. That helps overnight reviews and hotfix cycles.

For broader build strategy, see our software development consulting page and more guides on the Debugged Pro blog.

Security Features Agencies Should Brief Clients On

Clients will ask how secrets and scans work. Stick to verified product facts. Do not oversell.

Network secrets and the personal vault

Cloud environments support normal environment variables and network secrets. With network secrets, programs see placeholders. A proxy inserts the real value only for allow-listed HTTPS domains on port 443. That keeps tokens out of local process dumps and files during substitution.

Shared environments can request values each person supplies from a personal vault. Sharing the environment shares the requirement, not the personal credential. That is a clean model for client API keys that should never sit in a team-owned secret store.

Internet access and private networking

Internet access defaults to common package-manager registries such as npm, PyPI, crates.io, Maven, and related hosts. Teams can allow more domains or open wider access when the workflow needs it. Enterprise Agent Security policies can tighten this further.

For private APIs and internal registries, OpenAI currently supports Tailscale VPN. Auth keys should be reusable and ephemeral so new task VMs can join, then drop off. OIDC for short-lived cloud credentials is available by request for Enterprise workspaces.

Codex Security Cloud

Also announced at DevDay, Codex Security Cloud scans GitHub repos on demand, on a schedule, or when new commits land. Codex investigates findings, removes duplicates, and prepares fixes in the cloud while a laptop is closed. The set includes Daybreak Blue cybersecurity models without a separate application.

Agencies that already talk about agent governance may also want our OpenClaw Enterprise control plane guide for multi-agent ops beyond a single vendor stack.

Other DevDay Codex Updates Worth Tracking

Cloud environments were not the only Codex news. Agencies should note these nearby launches when planning client demos:

Refreshed Codex CLI with voice to start and direct tasks.
/agents view to delegate work and track multiple tasks.
Code review in ChatGPT desktop for GitHub pull requests and GitLab merge requests.
Automatic code reviews that run while users step away.
Decisions API (Luna) for real-time answers to predefined questions.
Agents API updates with computer use and Bedrock Managed Agents on AWS.

OpenAI's official docs for environments are at learn.chatgpt.com cloud environments. Use that page when a client asks for setup steps, not marketing copy.

Access, VM Sizes, and Hard Limits

Know the plan tiers before you promise capacity on a client call.

ChatGPT plan	vCPUs	Memory	Disk
Plus, Edu Plus	2	8 GiB	8 GiB
Pro, Business, Enterprise	4	16 GiB	32 GiB
Edu, Edu Pro	4	16 GiB	32 GiB

By default, task state is recoverable for about seven days after last use. Larger VMs are available for Enterprise through OpenAI's account team.

Current limitations to put in every SOW footnote

No GitLab repositories in cloud environments yet.
No self-hosted GitHub Enterprise Server yet.
No computer use or browser use inside cloud environments yet.
Personal local skills do not sync; repo-stored skills do.

If a client lives on GitLab today, keep local or CI-based agent flows until OpenAI ships support. Do not sell cloud environments as a full replacement for every repo host.

Agency Checklist for Codex Cloud Environments

Use this as a one-page brief for delivery leads in India and the USA.

Pick one pilot repo. Choose a GitHub project with clear install steps and low blast radius.
Publish one environment. Let Codex draft the install script and start skill, then review every command.
Separate run and edit rights. Share the environment with the workspace. Keep edit permission with leads or Enterprise admins.
Map secrets correctly. Use network secrets for HTTPS service tokens. Use env vars only when programs must read values directly.
Require personal vault entries for any credential that should stay with an individual, not the team store.
Lock the network policy. Start with package managers. Add domains only after a failed install proves the need.
Add Tailscale only when needed for private registries or internal APIs. Use reusable, ephemeral keys.
Turn on Security Cloud scans for the same GitHub repos on a schedule and on new commits.
Document the seven-day state window so clients commit or export work before tasks expire.
Write the limits into the client brief so GitLab and browser-use gaps do not become surprise blockers.

Key Takeaways

Codex cloud environments turn one-time project setup into a reusable, shareable baseline for cloud coding tasks.
Tasks keep running across phone, web, and desktop, even when a laptop sleeps.
Network secrets, personal vaults, Tailscale, and Agent Security give agencies real controls to brief clients on.
Codex Security Cloud adds scheduled GitHub scanning and cloud-prepared fixes, including Daybreak Blue models.
Plus VMs are smaller than Pro and Enterprise. GitLab, self-hosted GHES, and computer or browser use are still missing.
Agencies in India and the USA should pilot one approved environment per client stack before wider rollout.

FAQ: Codex Cloud Environments for Agencies

What are Codex cloud environments?
They are reusable setups for Codex Cloud tasks. Each holds repositories, dependencies, tools, and access settings. After you publish, every new task gets its own isolated VM from that setup.

Who can use Codex cloud environments?
ChatGPT Plus and Pro subscribers, plus Business, Enterprise, Edu, and Healthcare workspaces. Plus VMs are smaller than Pro and Enterprise tiers.

How do network secrets work in Codex cloud environments?
Programs see placeholders. A proxy inserts real values for allow-listed HTTPS domains. Personal vaults let each user supply their own values on a shared environment.

What are the current limits of Codex cloud environments?
No GitLab or self-hosted GitHub Enterprise Server yet. No computer or browser use in cloud environments yet. Task state lasts about seven days after last use by default.

Should agencies in India and the USA pilot now?
Yes, if clients use GitHub and an eligible ChatGPT plan. Start with one environment, clear secret rules, and documented limits before you scale.

About the Author

Nitin Kumar is Lead Project Coordinator at Debugged Pro. He helps software and IT agencies plan practical AI delivery for clients in India and the USA.