Codex Entry Points
About 5 minutes
Codex can be used from several entry points while staying connected to the same ChatGPT account. The important question is not just what the interface looks like. It is where the code is read, where commands run, and how much work you delegate to Codex.
OpenAI’s official materials describe Codex app, Codex CLI, Codex IDE extension, and Codex web as the main entry points. This page explains the practical differences for beginners.
Comparison
Section titled “Comparison”| Entry point | Execution location | Best fit | Watch out for |
|---|---|---|---|
| Codex CLI | Local terminal | File edits, tests, diffs, verification | Command permissions and working tree management |
| IDE extension | Local IDE | Fixing code in context, completion, review | IDE state and repository state must stay clear |
| Codex web | Cloud | GitHub issues, implementation, PRs, code review | Repository connection and cloud permissions |
| Codex app | Local or cloud | Multi-agent work, long-running tasks, workflow management | OS availability, limits, and workspace settings |
| ChatGPT mobile / web integration | ChatGPT surfaces | Lightweight instructions and review while away from the desk | Serious verification still needs CLI or web workflows |
Codex CLI
Section titled “Codex CLI”Codex CLI is the terminal entry point. It reads a local working tree and can edit files, run commands, execute tests, and inspect diffs.
Good fits:
- Small fixes in an existing repository
- Investigating test failures
- Documentation updates
- Refactoring
- Reviewing changes with
git diff
The CLI is powerful because it can run real commands in your local environment. Destructive commands, secrets, external writes, and production operations need clear approval rules.
Codex IDE Extension
Section titled “Codex IDE Extension”The Codex IDE extension brings Codex into IDEs such as VS Code, Cursor, and Windsurf. You can give instructions while looking at the active file, selection, errors, and project structure.
Good fits:
- Explaining the code you are editing
- Function-level changes
- Adding tests
- Fixing type errors
- Addressing review comments
The IDE extension is best for pair-programming style work where the developer keeps the code visible and Codex helps nearby.
Codex Web
Section titled “Codex Web”Codex web connects to GitHub repositories and delegates work in a cloud environment. It is useful for turning issues into implementation, creating diffs, and opening pull requests.
Good fits:
- Implementing from GitHub issues
- Multi-file changes
- Code review
- Pull request creation
- Parallel cloud tasks
For Codex web, GitHub connection and repository permissions matter. Confirm which repositories are accessible, whether PRs can be opened, and whether CI can run.
Codex App
Section titled “Codex App”The Codex app is an app-based entry point for managing agent work. OpenAI describes it as a command center for multi-agent and long-running coding workflows.
Good fits:
- Running multiple tasks in parallel
- Tracking long implementation work
- Switching between local and cloud workflows
- Using skills and workflows for repeatable tasks
OS availability, waitlists, plans, and limits can change. Check the official Codex page before rollout.
Local vs. Cloud Work
Section titled “Local vs. Cloud Work”| Perspective | Local work | Cloud work |
|---|---|---|
| Main entry points | CLI, IDE extension, local app workflows | Codex web, cloud app workflows |
| Code location | Your machine | Connected cloud environment |
| Permission model | Files, commands, network, approvals | GitHub permissions, cloud execution, PR rights |
| Best fit | Work you want to inspect locally | Delegation, parallel tasks, review, PRs |
How To Choose
Section titled “How To Choose”- Use CLI when you want to work in a local repository
- Use the IDE extension when you want help while reading and editing code
- Use Codex web when you want GitHub-connected cloud delegation
- Use the Codex app when you want to manage multi-agent or long-running work
- Use ChatGPT web or mobile for lightweight instructions and review
Summary
Section titled “Summary”- Codex connects to the same ChatGPT account across CLI, IDE, web, and app surfaces
- CLI and IDE are strongest for local work, while web is strongest for GitHub-connected cloud work
- The app is useful for multi-agent and longer workflows
- Choose an entry point by deciding where code lives, where commands run, how review works, and how PRs are created
See the references for the external specifications and background sources used on this page.[1][2]
References
Section titled “References”- OpenAI, Codex documentation
- OpenAI, OpenAI API documentation