OpenAI Codex app release – multi‑agent desktop interface for macOS and Windows

TL;DR

OpenAI released the Codex desktop app for macOS (with Windows support announced for March 2026), a unified command‑center that lets developers run, supervise, and collaborate with multiple AI agents in parallel, extend them with reusable skills, and automate repetitive tasks, all with built‑in security and configurable personalities.

Overview of the Codex app

The Codex app is positioned as a command center for agents that moves software development from single‑agent code suggestions to coordinated multi‑agent workflows spanning design, implementation, testing, and deployment. It integrates tightly with the existing Codex CLI and IDE extensions, importing session history and configuration so developers can adopt it without re‑configuring projects.

Parallel multi‑agent orchestration

  • Threaded project view: Each agent runs in its own thread, isolated from others, allowing developers to switch contexts without losing state.
  • Worktree support: Agents operate on isolated copies of the repository, preventing git conflicts while exploring divergent implementation paths.
  • In‑app diff review: Changes appear as diffs that can be commented on, opened in an external editor, or accepted/rejected directly.
  • Seamless onboarding: The app automatically pulls settings from the Codex CLI and IDE extensions, enabling immediate use on existing codebases.

Extending agents with skills

Codex is evolving from pure code generation to a skill‑driven execution engine that can invoke external tools, synthesize information, and perform non‑coding tasks.

  • Skill definition: A skill bundles instructions, resources, and scripts, letting Codex reliably connect to tools and follow team‑specific workflows.
  • Built‑in skill library: The app ships with a curated set of skills, including:
    • Implement designs – fetch assets from Figma and generate production‑ready UI code.
    • Manage projects – interact with Linear for bug triage and release tracking.
    • Deploy to the cloud – deploy to Cloudflare, Netlify, Render, or Vercel.
    • Generate images – use the GPT‑Image powered image‑generation skill.
    • OpenAI API reference – fetch up‑to‑date API docs.
    • Document handling – read, create, and edit PDF, spreadsheet, and DOCX files.
  • Custom skill creation: Developers can author new skills in the app, share them via repository commits, and invoke them from the app, CLI, or IDE extension.

Demonstration: a full‑stack game built by Codex

OpenAI showcased Codex building a 3D voxel kart racer (Voxel Velocity) using the develop‑web‑game and image‑generation skills. Starting from a single prompt, Codex consumed over 7 million tokens, acted as designer, developer, and QA tester, and produced a playable game after iterative reprompting. The prompt and skill definitions are publicly available for replication.

Automations for background work

  • Scheduled automations combine a prompt with optional skills and run on a user‑defined schedule.
  • Result queue: Completed automation outputs land in a review queue for developer approval.
  • Use cases at OpenAI: daily issue triage, CI failure summarization, release brief generation, bug detection, and periodic skill creation.

Personality customization

Developers can select between two interaction styles via the /personality command:

  • Terse – concise, execution‑focused responses.
  • Conversational – more engaging, empathetic dialogue. Both personalities retain identical functional capabilities.

Security‑by‑design

  • Native sandboxing: The app uses open‑source, system‑level sandboxing identical to the Codex CLI.
  • Default permissions: Agents may edit files only within their working folder/branch and perform cached web searches. Elevated actions (e.g., network access) require explicit user approval.
  • Configurable rules: Teams can define project‑level policies to auto‑grant specific elevated commands.

Availability, pricing, and rate limits

  • Release date: macOS launch on 2 February 2026; Windows support added on 4 March 2026.
  • Subscription access: Included for ChatGPT Plus, Pro, Business, Enterprise, and Edu plans. Free and Go users receive temporary access.
  • Rate limits: OpenAI doubled Codex rate limits for all paid tiers during the launch period; limits apply across the app, CLI, IDE, and cloud.
  • Credits: Usage is covered by the subscription, with optional additional credit purchases.

Roadmap and future directions

  • Platform expansion: Windows version already released; further OS support is implied.
  • Model upgrades: Continued improvements to GPT‑5.2‑Codex and faster inference.
  • Enhanced automation: Cloud‑based triggers will allow agents to run continuously, independent of a local machine.
  • Multi‑agent UX refinements: Ongoing feedback will drive smoother parallel workflow management.
  • Broader knowledge‑work ambition: While Codex excels at coding, OpenAI aims to generalize the agent framework to a wider range of technical and knowledge tasks.

Significance

The Codex app consolidates multi‑agent orchestration, skill‑based extensibility, and secure automation into a single desktop environment, lowering the barrier for developers to delegate complex, long‑running software projects to AI. By coupling these capabilities with generous rate limits and cross‑plan availability, OpenAI accelerates the adoption of AI‑augmented development across hobbyists and enterprises alike.

Sources