Codex

Codex is a software engineering agent from OpenAI that helps developers write features, fix bugs, answer codebase questions, and propose pull requests. Powered by codex-1 (a version of OpenAI o3 optimized for software engineering) and the newer GPT-5.2-Codex, it can work on multiple tasks in parallel while operating within secure, isolated environments. Since its initial launch in April 2025, Codex has evolved from a code generation tool into a comprehensive platform for orchestrating autonomous coding agents across entire software development lifecycles.

Native Desktop App for macOS

In February 2026, OpenAI released the Codex desktop app for macOS, representing a significant shift in how developers interact with AI coding agents. The app serves as a command center for managing multiple agents simultaneously, addressing a critical gap in existing development tools. While traditional IDEs and terminals excel at interactive programming, they were not designed for the emerging paradigm of multi-agent orchestration where tasks can span hours, days, or weeks.

The desktop app synchronizes with existing Codex CLI and IDE extension configurations, allowing developers to continue using their existing projects without reconfiguration. It includes a built-in terminal scoped to the current project or worktree for validating changes, running scripts, and performing Git operations without leaving the app.

Multi-Agent Orchestration with Worktrees

The app organizes agents in separate threads by project, enabling developers to switch between tasks without losing context. Built-in worktree support allows multiple agents to work on the same repository concurrently without conflicts. Each agent operates on an isolated copy of the code, letting developers explore different implementation paths without impacting their main codebase. As agents work, developers can check out changes locally or let them continue making progress without touching the local git state. Changes are only written to the main branch when explicitly approved.

Skills System

Skills extend Codex beyond code generation into broader development workflows. A skill packages instructions, resources, and optional scripts so Codex can reliably follow specific workflows and connect to external tools. Skills are created and managed through a dedicated interface in the app and can be shared across teams by checking them into repositories.

Available Skills Include:

• Fetching design context from Figma
• Managing projects in Linear
• Deploying applications to Cloudflare and Vercel
• Generating images using GPT Image
• Creating professional documents (PDF, spreadsheets, Word)
• Running evaluations and monitoring training runs
• Drafting documentation

Skills use progressive disclosure for context efficiency. At startup, Codex loads only the name and description of each available skill, then reads full instructions when a skill is invoked either explicitly by the user or automatically based on task matching.

Automations

Automations allow Codex to work in the background on scheduled tasks without requiring manual prompts. They combine instructions with optional skills and run on user-defined schedules. When an automation finishes, results land in a review queue for human review before implementation.

Common Automation Use Cases:

• Daily issue triage
• Finding and summarizing CI failures
• Generating daily release briefs
• Checking for bugs
• Monitoring alerts
• Creating reports on codebase changes

Automations operate using natural language instructions with AI reasoning rather than rigid scripts, allowing them to adapt to changing conditions. They run in dedicated background worktrees for Git repositories, keeping automation work isolated from active development.

Security and Sandboxing

Codex uses native, open-source, and configurable system-level sandboxing. By default, agents are limited to editing files in the folder or branch where they are working and use cached web search results. Commands requiring elevated permissions such as network access prompt for user approval. Teams can configure rules to allow certain commands to run automatically with elevated permissions.

During cloud task execution, Codex operates entirely within secure, isolated containers with disabled internet access (unless explicitly enabled), limiting interaction solely to code provided via GitHub repositories. The agent provides verifiable evidence of actions through citations of terminal logs and test outputs, enabling developers to trace each step taken during task completion.

Access and Pricing

Anyone with a ChatGPT Plus, Pro, Business, Enterprise, or Edu subscription can use Codex across all surfaces (CLI, web, IDE extension, and desktop app) using their ChatGPT login. Usage is included in subscriptions with the option to purchase additional credits. For a limited time, Codex access is also included with ChatGPT Free and Go tiers, and rate limits have been doubled for all paid plans.

Codex can be guided by AGENTS.md files placed within repositories for project-specific instructions and best practices.