OpenAI brings a macOS command center for multi-agent development
OpenAI released the Codex app for macOS, a desktop client designed to manage multiple coding agents concurrently, coordinate long-running work, and integrate with existing CLI and IDE workflows. The app arrives alongside temporary expanded access—Codex is included for ChatGPT Free and Go users for a limited time, and rate limits have been doubled across paid plans.
A workspace built for parallel agents
The Codex app positions itself as a multi-agent command center, where agents run in separate threads organized by project. Each thread preserves conversational and code-editing context, supports reviewing diffs and commenting in-line, and allows opening changes in a local editor for manual adjustments. The app also uses worktrees so multiple agents can operate on the same repo without clobbering a single local git state: agents work on isolated copies, and their changes can be checked out locally when needed.
Session state and configuration are picked up from the existing Codex CLI and IDE extension, enabling continuity across tools.
Skills: extending agents beyond code generation
Codex adds a concept of skills—bundled instructions, resources, and scripts that let agents interact with tools, run workflows, and complete tasks beyond simple code output. Skills are available in a dedicated interface inside the app and can be invoked explicitly or automatically based on task needs. OpenAI highlights a range of curated skills (the full list is in the open source skills repo), including integrations for:
OpenAI provided an example where Codex constructed a voxel kart racer using an image generation skill and a web game development skill, working independently across millions of tokens from a single initial prompt to handle design, development, and QA.
Skills can be checked into repositories and shared via team configuration; guidance on that workflow is available in the developer docs.
Automations and persistent background work
The app supports Automations—scheduled workflows that combine instructions and optional skills to run on a defined cadence. Results feed into a review queue for human oversight. OpenAI reports internal uses for recurring tasks such as daily issue triage, CI failure summaries, release briefs, and routine bug checks.
Configuration, personalities, and safety
Codex offers two agent personalities (a terse, pragmatic mode and a more conversational mode) selectable via the /personality command in the app, CLI, and IDE extension. Security and permissions are emphasized: the app uses native, open-source, configurable system-level sandboxing (the Codex repo is public) and, by default, limits agents to editing files in their working folder or branch and to cached web search. Agents must request permission for elevated actions such as network access; teams can define project rules to allow automatic elevation where appropriate (details are in the rules docs).
Availability and broader context
The Codex app is available starting today on macOS. Codex functionality is included for ChatGPT Plus, Pro, Business, Enterprise, and Edu subscribers using their ChatGPT login, with options to purchase additional credits if needed. For a limited period, Codex is also available to ChatGPT Free and Go users, and rate limits have been temporarily increased for paid plans.
Since GPT‑5.2‑Codex launched in mid‑December, overall Codex usage has doubled, and more than a million developers used Codex in the past month. OpenAI says future work includes a Windows client, expanded Automations with cloud triggers, and continued refinement of multi-agent workflows and inference performance.
Original source: https://openai.com/index/introducing-the-codex-app/