Codex
Codex is OpenAI's AI coding agent with a coding-specialized model, a two-layer workspace-plus-thread architecture, scheduled automations, skill packs, and sandboxed execution. It covers feature development, refactoring, and unattended operations across desktop, editor, and CLI.
Human verified · · Submit a correction
Best for professional developers and engineering teams that want to delegate development, refactoring, and ops work to an AI agent; not for non-programmers or users who only need simple code completion.
Verified facts
What is Codex
Codex is OpenAI's AI coding agent with a coding-specialized model, a two-layer workspace-plus-thread architecture, scheduled automations, skill packs, and sandboxed execution. It covers feature development, refactoring, and unattended operations across desktop, editor, and CLI.
Key features of Codex
- Describing tasks in natural language and letting Codex implement fixes and features with reviewable diffs
- Running parallel agents for complex refactoring and code migration to shorten delivery cycles
- Scheduling Automations for log inspection, issue triage, and other unattended operations
- Installing Skills so AI output follows team engineering conventions and code standards
Good for
- Two-layer workspace and thread architecture prevents context pollution across tasks
- Coding-specialized model with sandboxed cloud execution for safe, controlled runs
- Unified access via desktop app, editor plugins, and CLI under one ChatGPT account
Watch out
- The coding-specialized model is only available inside Codex, not on the main ChatGPT site
- The free tier is limited; heavy use effectively requires a Plus or Pro subscription
- Granting repository and system access to an AI agent requires careful permission setup
How to use Codex
- Download Codex from the official site and log in with your ChatGPT account
- Connect a GitHub repository or add a local project folder as a workspace
- Create a new thread, keeping one thread per task
- Describe the requirement in natural language and let Codex analyze and modify the code
- Review the changes, then commit to the branch or open a pull request
Who Codex is for
Difficulty: Intermediate
- Describing tasks in natural language and letting Codex implement fixes and features with reviewable diffs
- Running parallel agents for complex refactoring and code migration to shorten delivery cycles
- Scheduling Automations for log inspection, issue triage, and other unattended operations
- Installing Skills so AI output follows team engineering conventions and code standards
FAQ
Is Codex free?
Codex is freemium: the free tier covers basic features, while ChatGPT Plus and Pro subscribers get doubled allowances. Check OpenAI's official site for exact tier pricing.
How is Codex different from coding with ChatGPT?
Codex runs a coding-specialized model available only inside the Codex app, and wraps it in a full engineering workflow with workspaces, threads, sandboxed execution, and code review.
Can Codex work directly on my local projects?
Yes. Codex connects to local project folders or GitHub repositories and executes code in an isolated cloud sandbox, with multi-level permissions such as Local and Full access to control its reach.
Sources and verification
Sources: openai.com (opens in a new tab)
Verified: · Submit a correction →