Start Codex in a co-work
Press Open AI and pick Codex, or from a terminal:
$ chain open codex -t "Fix checkout double charge"
Codex starts in its own worktree on a branch for the task, with the board tools and the pre-tool hook already configured. It claims the paths it is about to change, works, and calls done when the tests pass.
Claims that survive a crash
A claim is only useful if it is released when the agent is finished, including when the agent stops unexpectedly. Chain releases a Codex session’s claims as soon as the session ends, so a crashed run never leaves a path locked for the rest of the team.
Codex and Claude Code on the same repo
Most teams do not standardise on one agent. Chain does not ask you to. A Codex session and a Claude Code session see the same claims, the same tasks and the same timeline, whether they run on the same Mac or on two laptops in different cities. When both want the same file, the second is blocked by the hook and told who holds it. The parallel agents guide explains why that beats hoping they stay apart.
Your OpenAI account stays yours
Codex runs on your Mac with your own OpenAI login or API key. Teammates never need it. Only the coordination goes through the relay, and on Enterprise you can host that relay yourself.