Run Claude Code and Codex on the same project

By Julien Abenhaim, founder of Castforge. Published 7 September 2026.

Yes, you can. They are two separate command-line tools with two separate sign-ins, so nothing stops you pointing both at one repository. The hard part is not launching them. It is keeping them from writing over each other, and knowing which one actually finished.

The short answer

In practice this is two terminals. In one you run claude, signed in to your Claude account. In the other you run codex exec, signed in with ChatGPT. Each tool keeps its own install, its own sign-in and its own session, and neither one knows the other exists. Both are looking at the same folder on disk, which is what makes the arrangement useful and also what makes it fragile.

What actually gets in the way

Three things go wrong, and they go wrong in this order.

Two terminals with no shared state. Whatever one agent decided, the other cannot see. If Claude picks a database column name and Codex picks a different one an hour later, nothing in either session contradicts the other. You find out at the merge.

Both agents editing the same files. Two edits to the same function in the same minute is not a hard problem for git, it is a hard problem for you, because you have to referee which intent was right. This is the failure that costs the most time, and it is the one people hit first.

No shared record. Come back an hour later and there is nothing to read but two scrollbacks, in two windows, in two different formats. Neither one is a summary of what the project now is.

Doing it by hand

The manual recipe works, and it is worth knowing even if you later use something to manage it. Install and sign in to each tool separately: Claude Code with a Claude subscription sign-in, Codex with a ChatGPT sign-in. Both install through npm and want Node.js 20 or newer.

Then give each agent a set of files the other one will not touch, and say so in the prompt. Commit between handoffs so the second agent starts from a clean tree rather than from the first one's half-finished edits. Keep the plan itself in a file inside the repository, a plain markdown file is enough, so that both agents can read the same thing instead of holding the plan only in a session neither shares.

Splitting the work so two agents do not collide

There are three splits that actually hold, and one of them will fit most tasks.

By file boundary. Agent A owns these paths, agent B owns those. The rule is checkable: if a diff touches a file outside the set, the split was wrong, not the agent.
By layer. One agent on the API, the other on the UI, with the contract between them written down first. This is the file split with a reason attached, which makes it easier to hold under pressure.
By turn. One plans, the other builds, never both at once. Slower, and the only split that is safe when the work genuinely does span the same files.

What this does not solve

Two agents editing the same file at the same time will still conflict. The only fence that reliably holds is disjoint files or sequential turns.
Each CLI keeps its own session and its own memory. Neither reads the other's history, so anything you want both to know has to live in the repository.
Running both spends against both plans. Two subscriptions, two quotas, two bills.
Each tool needs its own install and its own sign-in. There is no single sign-in that covers both.
Each provider's terms govern its own tool. Castforge drives the CLIs exactly as they ship and never extracts their sign-in tokens.

Doing it on a board instead

Castforge is a Windows desktop app that runs both as one team on a board. You choose which agent fills which role, Lead, Coder, Tester, Reviewer, Researcher and Designer, so your strongest model sits where it matters. Independent work runs in isolated lanes, each on its own branch, and lands back in order behind a review gate, so nothing merges unreviewed. Every card carries a reviewer and the plan lives in the project rather than in a scrollback.

It is Windows only today, in open beta, and free during early access. You can download it or read how the roles are assigned in your AI dev team.

AI coding agents on Windows. Which agents run on Windows today, how you sign in to each, and the traps in the way.
What is a reviewed AI dev team?. Roles, review gates, and what a gate is actually worth when it passes.
All Castforge walkthroughs.