OpenAI has released the Codex app for macOS, introducing a new way for developers to work with AI agents across complex and long-running software projects. The desktop application is designed to act as a central command interface where multiple agents can operate in parallel, allowing users to delegate tasks, monitor progress, and review results without losing context.
Since the launch of Codex earlier this year, developers have increasingly relied on AI agents not just for isolated code generation, but for broader responsibilities that span the full lifecycle of software development. As these tasks grow longer and more interdependent, OpenAI argues that traditional tools such as IDEs and terminals are no longer sufficient to supervise agent-driven workflows. The Codex app is intended to address that gap by giving developers direct visibility and control over multiple agents at once.
At the core of the app is a project-based structure that organizes agents into separate threads. Each agent runs in isolation, enabling parallel experimentation without risking conflicts in the underlying codebase. Developers can inspect changes, comment on diffs, and pull updates into their local environment when ready, while agents continue working independently in the background.
The app also introduces built-in support for worktrees, allowing several agents to operate on the same repository simultaneously. This approach lets developers explore alternative solutions or feature implementations without manually tracking how each path affects the main codebase. OpenAI says the app automatically synchronizes with existing Codex configurations from the CLI and IDE extensions, reducing setup friction for current users.
Beyond coding, Codex is being positioned as a more general execution engine through the use of skills. Skills bundle instructions, scripts, and resources so agents can reliably perform tasks such as deploying applications, generating documents, running evaluations, or gathering information. The Codex app includes tools for creating and managing these skills, enabling teams to standardize workflows and reuse them across projects.
OpenAI has also added Automations, which allow Codex to perform scheduled background tasks without direct supervision. These automations can handle repetitive but essential work such as issue triage, monitoring failures, or generating reports, with results delivered to a review queue for human oversight.
To accommodate different working styles, the Codex app allows developers to choose between distinct interaction modes, ranging from concise and execution-focused to more conversational and explanatory. OpenAI says this flexibility is intended to make agent collaboration feel natural while preserving consistent capabilities across modes.
The Codex app is available starting today on macOS and is included with ChatGPT Plus, Pro, Business, Enterprise, and Edu plans. For a limited time, OpenAI is also making Codex available to Free and Go users and increasing usage limits across paid tiers. OpenAI says it plans to expand availability to other platforms and continue refining multi-agent workflows based on developer feedback.
