Codex

M↓

Codex is OpenAI's cloud-based coding agent. It runs tasks in an isolated environment in the background, returning results when done — suited for longer-running or parallel coding tasks.

Codex is OpenAI's coding agent, released in 2025. It is distinct from the earlier Codex API (deprecated in 2023), which was the model behind the original GitHub Copilot. The current Codex is a cloud agent that runs inside a sandboxed environment with access to your repository, a terminal, and the internet.

How it works

You submit a task (prompt) — a feature to add, a bug to fix, a refactor to perform — and Codex works on it asynchronously in a cloud sandbox. It clones your repository, runs commands, edits files, and produces a pull request or patch when done. You can run multiple tasks in parallel. This makes it a different kind of tool: less interactive than a terminal agent, more like delegating to a background worker.

Agentic capabilities

Codex can read and modify code, run tests, install dependencies, and browse documentation. It is designed for tasks that can be fully specified upfront. It is less suited to exploratory, back-and-forth sessions where you iterate on a result in real time.

Access

Codex is available through ChatGPT Pro and higher subscription tiers and via the OpenAI API. Editor integrations are limited — the primary interface is the ChatGPT web app or the API.

Privacy

Code is sent to and executed on OpenAI's infrastructure. OpenAI states that API inputs are not used to train models by default. Review OpenAI's usage policies before using Codex with sensitive or proprietary codebases.

Found a tpyo?Edit

Cookies ahead

Our support chat tool "Intercom" would like to collect some more data on you. See the related link for more details.

Docs