How to Use Codex in VS Code: Complete Windows & Mac Setup Guide

Software teams are moving beyond simple autocomplete. Autocomplete helps with local syntax, but modern development problems are often larger than a single line: understanding an unfamiliar repository, tracing a regression across files, planning a migration, reviewing a pull request, or turning a product requirement into a small, testable implementation.

AI coding agents address that larger workflow. They can inspect project structure, reason about dependencies, propose an implementation plan, edit files, run approved commands, explain tradeoffs, and review their own output. That does not eliminate the developer. It changes where the developer spends attention: less on repetitive discovery and more on architecture, review, safety, and product judgment.

The value comes from structured workflows, not blind automation. A team that asks an AI agent to rewrite a system without review is creating operational risk. A team that uses an agent to analyze the codebase, propose a plan, make a scoped edit, and then pass through tests and pull request review is building a practical AI-assisted engineering system.

Codex is OpenAI's AI coding agent for writing, reviewing, and shipping code. It is designed for practical development work: understanding repositories, planning implementation steps, editing files, debugging errors, explaining systems, and supporting review.

Codex can be used across multiple surfaces, including the Codex app, Codex CLI, Codex IDE extension, and web or cloud workflows where available. This guide focuses on the VS Code and local development workflow because that is where many teams already keep their code, terminal, Git state, and validation commands.

The Codex CLI is an open-source command-line tool that can read, modify, and run code locally. In VS Code, that means you can use Codex from the integrated terminal even if the extension workflow is not available in your environment.

VS Code is where many developers already work. It contains the file explorer, editor, integrated terminal, Git view, extensions, local environment, and project-specific context. Bringing Codex into that environment reduces browser/editor switching and keeps AI assistance close to the actual development loop.

A practical VS Code layout is Explorer | Editor | Codex Panel, with the terminal at the bottom. If you are using the CLI fallback, use Explorer | Editor with a terminal running Codex. The goal is to keep code review, Git diff, terminal output, and Codex interaction visible at the same time.

This layout also helps standardize AI-assisted development workflows. Developers can ask Codex to inspect the repository, open relevant files, propose a plan, make a small edit, review the diff, run lint or build checks, and then decide whether the change belongs in a commit.

Codex is available across Free, Go, Plus, Pro, Business, Edu, and Enterprise plans, but usage limits vary by plan. Do not assume that Codex requires a paid-only plan, and do not assume every plan has the same capacity or administrative controls.

Higher-usage teams should review OpenAI's current pricing, admin controls, data handling options, workspace configuration, and usage limits before rolling Codex out broadly. Enterprise adoption is not only about whether the tool works; it is about whether the organization can govern how the tool is used.

On Windows, install Node.js from the official Node.js site if it is not already available. Then open a fresh VS Code terminal and verify both Node and npm:

On macOS, Homebrew is a practical option for many developer workstations:

If Node is already installed, simply run node -v and npm -v. If either command fails, restart the terminal, restart VS Code, or reinstall Node using your organization's approved package manager.

Install the Codex CLI globally with npm. This is the required command for the CLI workflow covered in this guide:

Then verify the installation:

Launch Codex from the VS Code integrated terminal:

If the codex command is not found, restart VS Code and the terminal, then check the npm global path. On managed machines, confirm that global npm installs are allowed by your IT policy.

Open the actual project folder, not only an individual file. Codex needs repository context: package files, configuration, components, tests, scripts, routes, and documentation. In VS Code, use File, then Open Folder.

Project-level context matters because coding agents infer patterns from the repository. A single file rarely contains enough information about architecture, naming conventions, test strategy, dependency policy, deployment assumptions, or UI patterns.

Where available, install the Codex IDE extension and open the Codex panel from VS Code. The extension workflow keeps the assistant close to the editor experience and can make it easier to discuss the files you are already reviewing.

If the Codex IDE extension is not available in your environment, use Codex CLI inside the VS Code integrated terminal. This is still a practical OpenAI Codex VS Code workflow because the assistant, source files, terminal output, and Git diff remain in one workspace.

Do not confuse the ChatGPT Work with Apps VS Code extension with the Codex IDE extension. ChatGPT Work with Apps is a different workflow for connecting ChatGPT with supported local applications. Codex is the coding agent workflow covered in this guide.

Launch Codex from VS Code or the integrated terminal, select sign in, complete browser authentication, return to VS Code, and confirm that Codex is connected. The exact screens may change over time, but the operating pattern is straightforward: authenticate, return to the project, and verify that the session is active.

Enterprise users may need workspace admin approval. If your organization manages ChatGPT access centrally, confirm which workspace, account, and policy settings apply before troubleshooting the local editor.

A weak first prompt is broad and ambiguous. It gives Codex too much room to guess:

Use a safer sequence instead:

This sequence is safer because it creates checkpoints. Codex first learns the repository, then proposes a plan, then implements a small change, then reviews the result. The developer can intervene before each meaningful side effect.

Enterprise teams should treat Codex as part of an engineering system. That means the organization defines where Codex can be used, what kinds of tasks are appropriate, which commands require review, how dependencies are approved, and what validation is required before a change can merge.

Approved lower-risk use cases often include documentation, internal tools, test scaffolding, refactoring with review, code explanation, migration planning, bug investigation, and repetitive implementation tasks. Higher-risk use cases include authentication logic, payment systems, production infrastructure, database migrations, security controls, permissions, and compliance-sensitive code.