From the archive
OpenAI launches a cloud coding agent preview
Codex starts repository tasks in separate cloud environments and returns changes for review through ChatGPT.
OpenAI launched a research preview of Codex on May 16, 2025. The cloud coding agent accepted tasks from ChatGPT and worked in separate environments preloaded with a software repository.
Repository work in the background
Tasks included implementing features, fixing bugs, answering questions about a codebase, and preparing pull requests. Codex could read and edit files and run the repository's tests, linters, and other commands. Several tasks could run independently, with results returned for review.
The launch used codex-1, a version of o3 trained for software engineering. The initial task environments had internet access disabled during execution.
Access during the preview
The rollout began with ChatGPT Pro, Team, and Enterprise accounts, with other plans scheduled to follow. Changes and test results remained available for inspection before integration. This was a background repository workflow, rather than only a chat conversation about code.