OpenAI Codex

OpenAI Codex is a coding agent for software development across the Codex CLI, IDE extension, Codex app, and cloud-based workflows. The open-source CLI runs locally from a selected directory and can inspect repositories, edit files, and run commands. Codex also supports code review, sandboxed local execution, GitHub workflows, MCP, subagents, and automation-oriented usage through non-interactive and cloud task modes.

Top features

• Local CLI agent: runs from the terminal, reads and changes code in the selected directory, and can execute local project commands.
• Open-source implementation: the CLI is published on GitHub and can be installed through documented package-manager paths.
• Multi-surface workflow: use Codex from the CLI, IDE extension, Codex app, and Codex Web, with cloud tasks for longer-running or remote work.
• Sandboxing and approvals: local commands run in constrained environments by default, with configurable sandbox modes and approval policies for file access, network access, and command escalation.
• Code review for GitHub: can review pull request diffs, follow repository guidance from AGENTS.md, respond to @codex review, and support automatic reviews when enabled.
• Repository guidance: AGENTS.md files provide project-specific instructions and review guidelines, with closer files applying to deeper parts of a repository.
• MCP support: Model Context Protocol can connect Codex to additional tools and context sources for repository and workflow tasks.

Use cases

• Inspect an unfamiliar codebase, trace request flows, and identify files relevant to a task.
• Implement scoped features, bug fixes, refactors, and code migrations with local edits and test runs.
• Run a second review pass on GitHub pull requests, focusing on serious issues and repository-specific review rules.
• Turn design references, screenshots, or Figma context into front-end changes with reviewable diffs.
• Add evaluation suites, improve test coverage, and run quality checks before committing changes.
• Automate recurring engineering workflows with non-interactive CLI runs, cloud tasks, GitHub comments, or custom skills.