Running two Claude Code accounts on one Mac

You have a work subscription and a personal one. Or a personal one and a client's. You sign into the second, and the first stops working. This is not a bug you can report. It is how the tool is built.

Why the second sign-in takes the first away

Claude Code stores its credentials in one place per machine, under your user account. Codex does the same. So does Antigravity. Each of them is written on the assumption that a machine belongs to one person and that person has one subscription.

That assumption held for a long time. It stopped holding the moment these tools got good enough that a single person wanted several of them running, on several plans, at once. The tool did not change. The way people use it did.

The practical result is simple. There is exactly one signed-in identity, and claude uses it. Sign in again and the stored credential is replaced, not added to. Nothing warns you, because from the tool's point of view nothing went wrong.

The four workarounds, and what each one costs

Sign out and back in

The obvious one, and the one nearly everyone starts with. It works, and it costs you about forty seconds each way plus whatever context you had in the session you killed.

The real cost is not the forty seconds. It is that you stop switching. You pick one account and leave the other one idle, because the friction is just high enough to not be worth it for a ten minute task.

A second macOS user account

Fast user switching gives each login its own home directory, so each one gets its own credential. This genuinely works and it is the closest thing to a supported answer.

It is also a whole desktop. Different terminal, different editor state, different everything, and you cannot see both at once. You are not running two accounts side by side. You are running one, twice, in alternation.

Containers or a VM

Give each account its own filesystem and the problem goes away. Docker, OrbStack, a Linux VM, all of them work.

What you pay is the mount dance, the performance hit on file watching, and the fact that the agent is now working on a copy of your project rather than your project. For an agent whose whole job is running commands against a real repository, that is a meaningful gap.

Fighting the config directory

You can point the tool at a different config directory per shell and swap between them. This is the power user answer and it does work, until a tool moves where it keeps things, or writes to an absolute path, or caches something in a place your override does not cover.

It is a workaround that needs maintenance, and it breaks quietly rather than loudly.

What running them at once actually requires

Every workaround above solves the credential problem and stops there. Running several accounts genuinely in parallel needs three things, not one.

1. 01Separate credentials per session. Each pane holds its own sign-in, and no session can read another's. This is the part the workarounds solve.
2. 02Separate working copies. Two agents editing one folder at the same time will overwrite each other's edits. They need somewhere of their own that is still the real repository, not a copy of it.
3. 03Somewhere to watch it. Three agents working in three places you cannot see is worse than one agent you can. Output has to land somewhere you can read at a glance.

Miss any of the three and you have a setup that technically runs and that you stop using within a week.

What we built instead

Nuvoa is a macOS app that runs each account as a pane in one window. Each pane holds its own sign-in, gets its own branch and working copy of the project, and streams its output where you can see it.

You can mix tools, so a Claude Code pane and a Codex pane and an Antigravity pane sit next to each other, each on a different account. Or run three Claude Code accounts and no others. The app does not care which, because a pane is just an account with somewhere to work.

The sidebar shows how much of its usage window each account has left, so you can tell before you start a round which accounts can take real work. An account that runs out does not lose its task. The work waits, and starts itself when the window comes back.

Two accounts really are separate, and neither one can reach the other's sign-in. That claim ships with a check that proves it, and the check runs on your own machine without a network or a model.

The honest limits

Nuvoa is macOS only. The work comes back as a branch for you to read, never as a merge, and nothing is ever pushed. An agent working on its own is running real commands on your machine while you are not watching it, and that is a trade you should make on purpose rather than by accident.

If you are already signing in and out several times a day, join the waitlist. One email, when there is a build.