As of , checked against the official docs.
Claude Code and Codex are independent tools, so you can use them together simply by running both in the same project. Three things make it work well: give each its own working directory with a git worktree so they never edit the same file at once, keep one AGENTS.md as both tools’ instructions, and settle how you will watch usage, approval requests and finished turns on one screen. A table below maps one tool’s commands to the other’s.
Why use Claude Code and Codex together?
Three reasons come up most.
- Their limits are separate. Claude Code draws from your Claude plan's limits and Codex from your ChatGPT plan's. When one hits its limit you can carry on in the other.
- Each can review the other's work. Have one agent read a change the other made and look for problems.
- Work can be split and run at once. One builds the feature while the other takes the tests or the docs.
This guide doesn't say which one is better. It covers what gets in the way when you run them side by side.
What should you watch for when both run in one project?
Give each its own working directory. Two agents editing the same files in the same folder overwrite each other. Make a second directory with a git worktree and run one agent in each.
git worktree add ../project-feature-a -b feature-a
cd ../project-feature-a
codex
Keep one instructions file. Codex reads AGENTS.md. Claude Code reads CLAUDE.md, and in a repository with an AGENTS.md and no CLAUDE.md it reads that as the project's instructions (Claude Code 2.1.277 or later). To keep both, put this one line in CLAUDE.md so both tools share the same file (Claude Code docs, Codex docs).
@AGENTS.md
Remember that approvals work differently. Claude Code decides what to ask through permission modes and allow rules, Codex through its approval policy and sandbox. A setting you loosen in one does not apply to the other.
How do the two tools' commands map to each other?
As of November 2026. If you know the left column, the right one does the same job.
| What you want | Claude Code | Codex |
|---|---|---|
| See remaining usage | /usage |
/status |
| See usage on the web | Settings > Usage on claude.ai | The Codex usage dashboard |
| Carry on a past session | claude --continue, claude --resume |
codex resume |
| Project instructions | CLAUDE.md (or AGENTS.md) |
AGENTS.md |
| Decide what it asks about | Permission modes (Shift+Tab), allow rules in /permissions |
approval_policy (on-request, never) with the sandbox, /permissions |
| A notification when done | Notification and Stop hooks | tui.notifications, notify |
How do you keep both in view on one screen?
Using both means more places to check: usage in two places, approval prompts in two terminals, and two places to see whether a turn has finished. There are three ways to handle it.
- Split the terminal and keep them side by side. The simplest. If both panes are always in sight, it is enough.
- Use each tool's own screen. Claude Code's agent view (
claude agents) manages Claude Code's background sessions only; Codex is watched in its own terminal or app. - Use an app that takes any agent. In Devalley, Claude Code sessions and Codex sessions are residents of the same village.
This is what the two look like together in Devalley. Each session is a resident that shows whether it is working, waiting for you or done. A resident brings an approval request to you with a "May I?" bubble, and you allow or deny it right there for Claude Code and Codex alike (for Claude Code, "Always allow" and its multiple-choice questions too). The usage card has a line each for Claude Code's and Codex's limits. Sessions in the Codex desktop app move in as residents too, showing whether they are working or done, and their card brings the Codex app forward. It is free on Windows and macOS.
There are things Devalley does not do. It doesn't hand one agent's work to the other or join their conversations. You decide who does what; Devalley lets you watch those sessions and answer them on one screen.
Frequently asked questions
When you use Claude Code and Codex together, are their limits combined?
No. Claude Code uses your Claude plan’s limits and Codex your ChatGPT plan’s, separately. That is why you can carry on in one when the other reaches its limit.
Can Codex and Claude Code run in the same repository at the same time?
Yes. Since two agents editing the same files in one folder overwrite each other, make a second directory with git worktree add and run one in each.
Do you have to maintain both CLAUDE.md and AGENTS.md?
No. Write the instructions in AGENTS.md and keep a single line, @AGENTS.md, in CLAUDE.md; both tools then read the same instructions.
Can you see Claude Code’s and Codex’s usage at once?
Yes. On macOS a menu-bar app such as CodexBar, and on Windows and macOS Devalley’s usage card, show both tools’ limits together.
Can you answer Codex’s approval requests without going to its terminal?
In Devalley a resident brings Codex’s approval request to you, and you allow or deny it there. Nothing is allowed without your click.