How I Run Two Claude Accounts as One

inulute — Wed, 27 May 2026 13:00:00 +0000

Right! When Claude is finally getting it - your codebase, your weird architectural choices, that one bug you've been chasing for three days, BOOM! it stops. In between!

Rate limit exceeded. Please try again later.

Later!? Like am I supposed to just pause my brain and come back in another five hours?

I'm a developer. I don't have five hours. I have right now.

The problem is the wait

Five hours. That's what gets me.

And those five hours? They are never convenient. The limit doesn't wait for a clean stopping point. It hits mid-bug, mid-thought, right when the fix was one prompt away. It gets worse when there's a deadline on the other end: the PR due tonight, the demo tomorrow morning, the feature you already promised. The rate limit does not care about any of that or your credibility.

"Come back later" also promises that later is useful. Hit the limit at 3pm and the reset lands after you've logged off for the day. Hit it late and it resets at some hour you'll sleep through. The window opens when you're not there to use it.

You can't just go and do something else. You feel very helpless. When you're deep in a Claude Code session, that session is the work. There's no parallel task sitting there waiting for you. You're not freed for the next five hours. You're blocked for those five hours.

That is the real cost. Not the limit, but the dead time it drops you into. And usually right when you can least afford it.

What I tried first, and why none of it worked?

Option One: Upgrade to Max.

Anthropic wants you on Claude Max. That's $100 a month, with a $200/month tier above it.

But Max is built for people running Claude Code 24/7, burning tokens like they are free. That's clearly not me. That's probably not you either.

I didn't need a lot more. I needed a little more. Just enough to finish a session without getting cut off. Paying 5x for headroom?? That I'd never fully use?? It felt wrong.

Option Two: the API.

Pay per token, no hard limits. Sounds clean, until you do heavy Claude Code sessions and watch the bill spiral well past $100/month. Out.

Option Three: Do the math.

Claude Pro is $20/month. Two Pro accounts is $40/month. Combined, that comfortably covers that "just a little more" problem for most developers.

$40 versus $100 : For someone who just needs to finish a session, that choice makes itself.

So I got a second account. And then, immediately hit a new problem.

There was no clean way to actually use these two accounts together. Every time the first one ran out, I did it by hand: log out, log in with the other account, restart the session, wait for everything to reload.

Every. Single. Time.

It's not that any of these single steps are that hard. It's just that the whole thing yanks you out of the problem you were actually trying to solve. One second you're chasing a bug, the next you're doing the account admin. I started calling it the "Logout Dance". Two Pro accounts fixed the wait. They did nothing for the dance.

So I built the missing piece.

I built cux

cux is a CLI wrapper around Claude Code. You run cux instead of claude. From your side, that's the whole thing. Run cux that's it.

Under the hood, it watches your session, detects when a rate limit is approaching, and rotates to your next account automatically, with --resume so your conversation picks up exactly where it left off.

No logout dance. No babysitting. No getting yanked out of the problem at 2am.

You won't even notice it switched. No restart. No reload. Claude just... keeps going.

Why was this trickier than killing a process?

The naive version of cux is easy: detect the rate limit, kill Claude, swap accounts, restart. I wrote that first. It didn't work.

The problem is timing. A rate limit doesn't wait for a polite moment. It lands whenever it lands, often while Claude is mid-response, halfway through editing a file or working through a step. Force-kill the process right then and you don't get a clean switch. You get a session interrupted in the middle of an action, left in a state nobody asked for.

The fix was to stop fighting Claude and work with it. Instead of killing the process the instant a limit shows up, cux waits for a clean boundary. It hooks into Claude's own Stop signal and rotates accounts then.

That's the part I'm weirdly proud of. The difference between "I wrote a script" and "I built a tool" is usually this exact kind of edge case.

Is this even allowed?

Fair question, and worth answering straight.

cux isn't a bypass and it isn't a hack. It doesn't spoof anything or hide anything. Anthropic sees every request exactly as it's made. All cux does is let you use two subscriptions you already pay for, together, without the manual friction.

To be honest about it: automated account rotation lives in a gray area of how subscriptions are meant to be used and that's worth knowing before you adopt it. cux doesn't do anything sneaky, but it's a workflow tool, not an official feature. Use it with that understanding.

Getting started

Before you install, you'll need:

Claude Code already set up and working
Two Claude Pro accounts
Node.js installed (only if you're using the npm method below)

There are two ways to install cux.

Option 1 — npm

npm install -g @inulute/cux

Option 2 — shell installer

Works on Linux, macOS, WSL and Git Bash on Windows.

curl -fsSL https://raw.githubusercontent.com/inulute/cux/main/scripts/install.sh | sh

Once it's installed, set up your accounts:

cux setup
cux add # add account 1
# logout and login with account 2
cux add # add account 2
cux status # verify accounts

cux # launch via "cux" instead of "claude"

There are three rotation strategies built in:

drain: use one account fully before switching
balanced: always pick the freshest account
manual: switch yourself with /switch Pick whatever fits your workflow. Full setup and options live at cux.inulute.com/docs.

Wrapping up

cux exists for one reason: a rate limit shouldn't stop you working. Two Pro accounts, rotated automatically, mid-session, with no logout dance. That's the entire idea.

It's still young. There are rough edges I haven't found yet, and probably some you will. If that happens, I genuinely want to hear it. Try it, break it, tell me what's wrong with it.

And if it saves you even one session from the five-hour wait, that's enough for me.

GitHub github.com/inulute/cux · npm @inulute/cux · Website cux.inulute.com

Drop a ⭐ if it saves you from the logout dance.

DEV Community: inulute