adamallcock/codex-chatgpt-control

What it solves

It provides a structured way for AI agents (specifically those using the Codex framework) to delegate tasks to the visible ChatGPT web interface. This prevents users from having to manually move work between local execution environments and the ChatGPT web UI for conversational review, research, or long-form deliverables.

How it works

The SDK acts as a bridge between an agent and the chatgpt.com website. It does not use a private API; instead, it drives the browser through a compatible Codex/browser bridge, interacting with the visible UI controls. It can detect whether the user is in the "Chat" or "Work" experience, inspect and apply configuration settings (like model or effort levels), manage threads, upload files via visible controls, and download generated artifacts and images.

Who it’s for

Developers building Codex-style agents who want their agents to be able to interact with the real ChatGPT web product that the user can see and monitor.

Highlights

• Visible-Session Only: Operates through the browser UI, ensuring the workflow remains user-directed and visible.
• Experience Detection: Distinguishes between ChatGPT Chat and Work experiences and verifies the resulting UI state.
• Artifact Management: Supports uploading local files and downloading generated files or image-only artifacts.
• Transactional Operations: Includes a durable operation ID system to prevent duplicate tasks and allow recovery from partial results.
• Multi-Language Support: Available as both a Node.js runtime and a Python parity client.