OpenAI launches Codex AI coding app for Windows with native sandboxing and multi-agent support

OpenAI Brings Codex to Windows After macOS Debut

OpenAI has officially launched its Codex AI coding app for Windows, expanding access to developers using Microsoft's operating system roughly a month after releasing the macOS version 1

1

2

2

. The native Windows application represents a proper platform-specific build rather than a simple port, designed from the ground up with Windows developers in mind 3

3

. Available through the Microsoft Store or directly from OpenAI's website, the software engineering agent serves as a command center for agentic software development, allowing users to generate code through natural-language prompts 4

4

5

5

.

Multiple AI Coding Agents Work in Parallel

The Windows version enables developers to coordinate multiple AI coding agents to work on the same task simultaneously, a capability OpenAI upgraded in early February 1

1

. Users can direct agents on complex coding projects ranging from generating full applications to debugging large codebases, with the agents powered by specialized versions of ChatGPT and large language models (LLMs) fine-tuned for coding 5

5

. Codex can translate natural-language prompts into code across multiple programming languages including Python, JavaScript, and Go, while also proposing pull requests, fixing bugs, and generating tests 5

5

. The application lets users manage complex projects by planning out actions before executing them, offering detailed roadmaps that developers can review and adjust before giving approval 2

2

.

Native Features and Platform-Specific Tools

OpenAI incorporated native sandboxing into the Windows app, running natively using PowerShell inside a dedicated Windows sandbox to help developers feel at home 1

1

3

3

. The platform includes a dedicated Skills section that bundles together instructions, resources, and scripts the software can use to connect agents to specific tools and workflows, including a WinUI skill specifically for Windows app developers 3

3

. Automations streamline repetitive tasks like bug testing, allowing Codex to work in the background without constant supervision 3

3

. Work Trees enable users to run multiple independent tasks within the same project simultaneously, keeping the work of AI agents in a sandbox before deployment to a live production environment 2

2

3

3

.

GitHub Integration and Cross-Platform Workflow

Developers can connect Codex to local directories and GitHub repositories, allowing agents to check out branches of GitHub projects before making modifications 2

2

. For those who prefer working in Windows Subsystem for Linux (WSL), the application supports switching both the coding agent and integrated terminal to run inside WSL 3

3

. Session history is saved to an OpenAI account, meaning developers can start coding on Mac and then move to Windows without losing their work, enabling seamless cross-platform continuity 1

1

4

4

. A left-hand column lets users keep tabs on multiple Codex chats at once, with notifications appearing when pending agent actions need approval 2

2

.

Subscription Tiers and Usage Considerations

Codex is accessible to users with ChatGPT Free, Go, Plus, and Pro subscriptions, requiring sign-in with an existing ChatGPT subscription or an API key 1

1

3

3

. While free and Go tier users can access basic features, those on Plus ($20 a month), Pro ($200 a month), Business, Enterprise, and Edu plans enjoy higher rate limits and premium capabilities 2

2

5

5

. However, developers should note that high usage can quickly consume token allowances even with paid plans, given the rate at which AI coding agents burn through tokens 2

2

. Users can adjust agent autonomy levels, from requiring approval before every command to fully autonomous operation, though the latter approach can be both unpredictable and expensive 2

2

.

Competition in Agentic AI Coding Space

Codex enters a growing field of agentic AI coding tools, competing with Anthropic's Claude Code app and Google's Antigravity 2

2

. The Codex model, first unveiled in April 2025, already powers tools such as GitHub Copilot and the Codex CLI 5

5

. Originally released as an open-source command-line tool last April before the macOS-native desktop app landed in early February, Windows users had to wait several additional weeks for their own version 2

2

. This release signals OpenAI's commitment to supporting developers across different platforms, though the rapid token consumption and usage limits suggest that project planning and careful agent management will be critical for developers looking to integrate these tools into their workflows without incurring significant costs.