The short answer
Codex is OpenAI’s coding agent. It can inspect code, make changes, run commands, and automate repeatable work. OpenAI lists it on the ChatGPT desktop, mobile, and web apps, in a Codex CLI, in an IDE extension, and in Codex Cloud. Teams should pair it with an AGENTS.md file, permission settings, and human review.
What is Codex?
Codex is OpenAI’s agentic coding tool. OpenAI’s documentation describes the CLI as a way to inspect code, make changes, run commands, and automate repeatable work without leaving your terminal. As with other coding agents, you describe an outcome and it works through the steps while you review.
Codex has also become more central to the ChatGPT product. According to The Next Web’s report on the July 2026 launch of ChatGPT Work, the Codex desktop app became the new ChatGPT desktop app, and ChatGPT Work is built on Codex technology. Product naming moves quickly, so confirm current names in OpenAI’s documentation.
Where can you use Codex?
OpenAI’s documentation lists these surfaces: the ChatGPT desktop app, the ChatGPT mobile app, ChatGPT on the web, the Codex CLI, the Codex IDE extension, and Codex Cloud, which has its own documentation along with local environments.
| Surface | Good for |
|---|---|
| Codex CLI | Terminal-based daily work and scripting |
| IDE extension | Working inside your editor |
| Desktop app | A graphical home for sessions |
| Codex Cloud | Tasks that run in the cloud instead of on your machine |
Search interest reflects the shift. Bing Webmaster Tools shows strong impression volume for "codex app" and "codex download" over the past three months, which suggests people are looking for the installable app and CLI, not only the concept.
How do you get started with the Codex CLI?
OpenAI’s page gives a standalone installer for macOS and Linux and lists Windows, npm, and Homebrew options. The documented macOS and Linux command is:
curl -fsSL https://chatgpt.com/codex/install.sh | sh
Then run codex in a project directory. On first run, choose "Sign in with ChatGPT" or another available sign-in method. The /init command creates an AGENTS.md file with instructions for Codex, and /permissions lets you choose when Codex can edit files or run commands without asking and inspect the active sandbox.
Commands change. Confirm them in the official CLI documentation before giving them to a team.
What can you extend Codex with?
Per the documentation, the codex mcp command adds local or remote MCP servers, authenticates when needed, and lets you inspect available tools before Codex uses them. You can package repeatable instructions as skills, and add plugins to connect Codex to your team’s tools and data.
These mirror the building blocks on the Claude Code side, which is why the training concepts transfer: instruction file, skills, tool connections, permissions. See our guide to Claude Code hooks, skills, and subagents for the parallel.
What should a team do before adopting Codex?
Do four things first:
- Decide which repositories and environments are in scope.
- Write an AGENTS.md with commands, conventions, and guardrails.
- Set permission and approval rules, and understand the sandbox.
- Require human review of every agent-authored change.
Then pilot with one team, measure against a baseline, and train on your own repository. For the full safety checklist, see Codex team training and safety.
How do you install and start Codex?
Search interest in "Codex download", "Codex install", and "Codex CLI" is high. According to OpenAI’s Codex documentation, the CLI can be installed with an install script (curl -fsSL https://chatgpt.com/codex/install.sh | sh), then started by running codex in a project directory. Inside the CLI, the /init command creates an AGENTS.md file, /permissions adjusts what Codex may do, and codex mcp manages MCP servers.
OpenAI has also described a Codex desktop app. Because Codex plans, apps, and names have been changing (including its relationship to ChatGPT Work), confirm current download steps and plan eligibility in OpenAI’s documentation before you instruct a team. Always read an install script before piping it to a shell, and ask IT how your organization wants developer tools installed.
A safe first session: start in a throwaway or low-risk repository, ask Codex to explain code, then to add tests, and review every change.
What do Codex, Codex CLI, and the Codex app mean?
People use "Codex" for several things: the underlying coding agent, the command-line tool, and OpenAI’s desktop app. For a training program, clarify which one you approve. Pick the surface your engineers will use, document the install and update path, and put your conventions in an AGENTS.md file so the agent follows them. For comparisons with Claude Code, see Codex vs Claude Code.
Frequently asked questions
Is Codex the same as ChatGPT?
Codex is OpenAI’s coding agent. OpenAI lists it on ChatGPT surfaces, and reporting says the Codex desktop app became the new ChatGPT desktop app. Check current documentation for exact packaging.
Is Codex free?
Plan availability and usage metering change. Check OpenAI’s current plan documentation. We do not quote prices because they change often.
Does Codex work with MCP?
Yes. OpenAI’s CLI documentation describes a command for adding and authenticating MCP servers.
How is Codex different from Claude Code?
They are similar in purpose and differ in ecosystem and controls. See our comparison.
How do I install Codex CLI?
OpenAI’s documentation provides an install script and then you run codex in your project. Verify the current command in the official docs and review scripts before running them.
What does /init do in Codex?
According to OpenAI’s documentation, /init creates an AGENTS.md file in your project to hold instructions for the agent.
Sources and further reading
Product features, plans, and limits change often. This guide reflects the vendor documentation listed above as of November 2026. Check the vendor’s current documentation before you make purchasing or policy decisions. TechStack Consulting is an independent training firm and is not affiliated with Anthropic or OpenAI.