We are rolling out Codex in ChatGPT to users on Plus, Pro, Team and Enterprise plans. We plan to start rolling out to Edu in the coming weeks.
For the full technical user guide, see https://platform.openai.com/docs/codex.
For information about the Codex CLI / Sign in with ChatGPT see Codex CLI and Sign in with ChatGPT.
For ChatGPT Enterprise admins see our Admin getting started guide.
Overview
What is Codex in ChatGPT?
Codex is OpenAI’s cloud-based software engineering agent—purpose-built to support professional developers by securely connecting to your codebase and automating tasks like bug fixing, feature development, test generation, and more. It runs tasks in parallel in the cloud, so engineers stay focused while Codex handles the rest.
How workspace admins can turn Codex ON
In your ChatGPT workspace settings at https://chatgpt.com/admin/settings under Codex toggle Allow members to use Codex to ON.
Security / training / privacy
How can Enterprise admins feel confident about security when they enable Codex workspace-wide during the research preview?
Every Codex task runs in its own ephemeral container, created using your environment settings. During container setup, internet access is available so that you can install dependencies.
After the setup phase, the Codex agent gets access to the container and starts working on its task. As the workspace admin, you control whether the container has network access while the agent does its task. If you choose to disable network access while the agent does its task, all outbound traffic is blocked.
When the agent is done, you are presented with the diff produced by the agent, for review before merging the changes into your codebase. Additionally, you can inspect the agent's work log. The work log contains every command run by the agent.
ChatGPT access to Codex is turned OFF unless a workspace admin manually toggles Codex ON for the workspace.
The GitHub token is scoped only to the repo the admin connects.
Do you train your models based on my usage of Codex?
For ChatGPT Team, ChatGPT Enterprise, and ChatGPT Edu users, we do not train models on your Codex content, consistent with our Enterprise privacy commitments.
For users on all other plans, model training depends on the "Improve the model for everyone" setting in ChatGPT. If this setting is on, your tasks can be used to improve our models.
There is an additional setting in Codex that allows you to enable training on your complete environments, including the connected repos. It is off by default for all environments.