OpenAI Codex release notes: Windows gets Computer Use and remote control

The latest OpenAI Codex release notes matter if you want agents to work in real work environments, not only inside a code folder. With Codex 26.527, OpenAI moves more Computer Use and remote-control support onto Windows. A Windows PC can now become a host for tasks where the agent needs to see, click and type in a graphical interface.

OpenAI Codex is OpenAI's coding agent across app, IDE, CLI, and cloud. Computer Use means Codex can see and operate graphical apps on the computer. A sandbox is the bounded environment where the agent's commands can run. An approval gate is the point where a human approves an action before the agent continues.

OpenAI Codex release notes: what 26.527 changes on Windows

OpenAI lists Codex 26.527 in the Codex changelog on May 29, 2026. The main change is not a new text model. It is that Windows becomes a more complete work surface for agent tasks.

• Computer Use now works on Windows, so Codex can operate Windows apps by seeing, clicking and typing in the foreground.
• Remote control supports Windows devices. OpenAI says work can be started from ChatGPT on iOS or Android, or from a Mac running Codex, and checked remotely.
• The Profile section now shows profile details, usage stats and token activity.
• Search across past Codex app threads now includes conversation content and Git branch names.
• Local projects and worktree setups get better thread coordination, including separate background threads when explicitly requested.

Source: OpenAI Codex changelog: Computer use and mobile access on Windows

One important note for Swedish and European readers: OpenAI's Computer Use documentation says the feature is available on macOS and Windows at launch, except in the EEA, the United Kingdom and Switzerland. For Swedish teams, this may mean planning and monitoring first, then hands-on testing when your workspace and region have support.

Source: OpenAI Codex app: Computer Use

Why Windows support matters for practical teams

Plenty of business workflows still live in desktop apps, browser windows, accounting tools, admin panels and old software without a clean API. Giving an agent a repository is not always enough. Sometimes it has to check whether the real window behaves correctly.

The new Windows support makes those tests more realistic for organizations that already work on Windows. It also makes the control question more concrete. When Codex can move the pointer and type in the foreground, the team has to decide which apps are allowed, which steps need approval and how the result will be reviewed afterward.

For Hammer readers, this belongs in a Tool Forge project: build one small, reviewable routine before connecting the agent to more systems. Use separate test accounts, keep sandbox boundaries, put secrets in a secret manager or environment variables, use scoped permissions and keep logs, so the work can be reviewed later.

How to test the OpenAI Codex change without mixing up the steps

Human step first: if your region and workspace support it, open the Codex app on Windows and install Computer Use through Codex settings > Computer Use > Install. Keep the target app visible, because OpenAI says Computer Use on Windows runs in the foreground and takes over pointer and keyboard input while the task runs.

If you want to monitor work from your phone, mobile setup starts in the Codex app on the host, not in the Codex CLI or IDE Extension. OpenAI describes the flow as Set up Codex mobile in the sidebar, QR scanning on the phone and connection review under Settings > Connections.

Source: Computer Use in the Codex app and Remote connections in Codex

Short example: use the new Codex feature

Once Computer Use is installed and the right app is already open, give Codex a short, bounded task.

Use @Computer to inspect the Windows app flow I have opened. Reproduce the visible demo path, write down each UI step you take, and stop with a short fix plan before changing files. Ask for approval before using any credentials, changing settings, or editing code.

Good output should:

• Name the app or window Codex actually saw,
• Separate UI observations from proposed code changes,
• Ask for approval before files, settings or logins change,
• End with a short review list: steps tested, remaining risks and the next human decision.

What we are watching next in the Codex changelog

This does not replace Codex CLI release notes. It sits beside them. The CLI is still better for code, tests and repeatable commands. Computer Use is for the last visual checks where a window, button or older tool decides whether the workflow actually works.

For Swedish organizations, the next question is simple: when regional support arrives, which two or three GUI workflows are worth letting Codex observe first? Start there. Not with the whole operation.