Field note 03

Codex Joins the ChatGPT Desktop App

Codex now lives inside the ChatGPT desktop app, bringing repository work, review, Computer Use, and plugins into one surface.

By Codex Rules EditorialJuly 10, 20266 minute read

On July 9, Codex moved into the ChatGPT desktop app. Existing Codex users keep their projects, settings, and workflows, but the surrounding surface now includes more of the work that happens before and after an agent writes code.

That is the meaningful part of the change. The agent is no longer presented as a separate coding destination. It sits beside the conversations, connected tools, and review workflows people already use.

What is confirmed

According to the official changelog, the desktop app can edit Markdown and code directly with inline annotations. GitHub pull request review appears in the sidebar. Projects can work across repositories. Computer Use has been made faster, and plugin management moved into Settings.

The plugin directory also replaced the older app directory language. OpenAI's help documentation says a plugin may include skills, apps, and app templates. Existing app connections remain in place. Installing or using a plugin still depends on plan, workspace role, surface, and region.

The permission model deserves attention. Plugins inherit the permissions of the apps they contain. OpenAI recommends reviewing the capabilities, authentication requirements, read and write access, and approval behavior before enabling one. A read-only first rollout is a sensible default.

The questions users are asking

The Codex team held a Reddit AMA on July 10, with an official OpenAI Developers post on X used as proof. The question thread is more useful as a map of user concerns than as a product specification. People asked about model selection, usage limits, checkpoints, connectors, sandboxing, user interface details, and how Codex fits with their existing tools.

Those questions share a theme: once an agent becomes part of daily work, raw capability is not enough. People want visibility into state, cost, permissions, recovery, and control.

Community questions are not confirmed roadmap commitments. They are still valuable signals about where friction lives.

What to check after the move

If you already use Codex, the migration is a good moment to audit the surrounding setup:

Confirm each project points to the intended repository and branch.
Review connected apps and installed plugins.
Check which plugins have write access.
Revisit approval settings for network, filesystem, and external actions.
Open one existing task and verify that its history and local context look right.
Try a small pull request review before relying on the new sidebar for a large one.

The point is not to distrust the migration. It is to use a visible product change as a prompt to review invisible permissions.

My take

Putting Codex inside ChatGPT makes the product easier to discover, but its deeper value is continuity. A conversation can become a plan. A plan can become repository work. A diff can become a review. A plugin can connect the task to another system.

That continuity also makes boundaries more important. When chat, code, browser control, and external apps share one surface, users need clear signals about which capability is active and what it can change.

The winning interface will not be the one that hides all complexity. It will be the one that makes the right complexity visible at the moment a decision is required.

A rule worth copying

When a coding agent gains a new surface or integration, recheck repository scope, external permissions, approval behavior, and recovery before using it on critical work.