Hot swapping: chats that move when an account runs out

Hot swapping AI accounts, the way Shuriken does it: your running chats move to another of your accounts when one nears its limit. What happens at 99%.

99% beside two real screenshots: Shuriken's fuel gauge: what Claude and Codex have left this week, and Shuriken's switch settings: switch automatically at 99%
Contents

Your Claude account hits 99% of its 5-hour window halfway through a refactor. Shuriken moves the running chat to another of your accounts or sources, one that still has room, and the agent keeps going in the same chat. We call that a hot swap. Nobody logs out, and nobody pastes the context back in.

In the app the feature is still called switching. We wanted a name people would recognise, and "hot swapping AI accounts" says it better.

Hot swapping AI accounts: why the name#

In hardware, a hot swap is replacing a part while the machine keeps running: a disk, a power supply, a fan. Nobody shuts the server down. That's the feeling we were after.

Before Shuriken, we did it cold, every time a limit hit:

  1. Stop the agent.
  2. /logout in Claude Code, then /login with the other account.
  3. Find the session again with claude --resume, or start a new one and explain the task from scratch.

Five hours later, the same steps the other way round. We built Shuriken so that happens without us, and the name came with it.

What actually happens at 99%#

Every switching setting is under Switch settings at the top of Sources. The defaults are what most people run.

A swap happens when the account in use reaches your switch point, or when the provider reports a limit, whichever comes first. The switch point is 99% of what the app shows unless you change it, anywhere from 50% to 100%. For Claude it applies to both the 5-hour window and the weekly one.

Shuriken's Switch settings: Switch automatically on, Switch at 99%, How set to When one runs out, Start unused session limits on, Switch to API keys off, and the five rules for how the next source is picked.
The switch settings: switch automatically, at 99% unless you change it.

Your chats move to the source with the most room. Every rule behind that pick, including Distribute mode and what happens when two accounts are close, is in How Shuriken picks the next account.

How a chat moves depends on the agent.

  • A Claude Code chat moves on its next request, together with its subagents and workflows. That request is replayed on the new account, so a long task doesn't stop halfway.
  • Other agents (Codex, Gemini CLI and the rest) move when their current turn ends. The next turn starts on the new source.
  • The swap reaches every device at once. If you're following the chat from your phone, you see the same thing as your desk.

Each swap leaves a Switched to … card in the chat saying which account took over and why. No pop-up, nothing to click.

If every source is out, the chat shows one pause card ("Paused: no Claude source has usage left") and carries on by itself once a source has room again. Shuriken doesn't move a chat onto an API key from its own provider unless you turned on Switch to API keys, which is off by default. With it on, keys take chats only once every subscription is out, and a spend cap per fallback run ($25 unless you change it) stops the run. One exception: if you added an endpoint for that agent that serves another company's models, Kimi or GLM for example, the chat can go on there between turns once every source of its own company is out.

The one cost of a swap#

Better you hear it here than spot it on a bill. Claude keeps a prompt cache, so each new request re-reads the conversation at the cheaper cached rate. That cache doesn't travel between accounts: Anthropic's docs say caches are isolated between organizations (Claude Code docs), and a personal Claude account is an organization of its own.

The first request after a swap therefore reads the whole conversation with no cache hits, as a model switch would. From the second request on, the new account has its own cache. On a long chat you pay for one uncached turn, and the agent doesn't stop.

What hot swapping isn't#

It isn't a way to get more than you pay for. Shuriken only uses the plans and keys you add as sources. We never sell or pool accounts, and we never suggest buying extra ones to get around a limit.

Settings › Sources: a fuel gauge with about 142 of 258 fuel left this week across Claude and Codex, and five sources with their limits: two Claude subscriptions, an Anthropic API key with $28.00 of $100.00 in credits, a Codex plan and a Gemini plan.
A hot swap only moves chats between the sources you added yourself.

Use accounts you own and are entitled to use. Providers' terms may not allow sharing a subscription sign-in, and it can put your account at risk. If you're unsure whether two Claude accounts are fine for you, start with what's known about multiple Claude accounts.

Who else hot-swaps#

We're not the only ones using the word, and that's fine. Orca, a free and open-source agent IDE, swaps Claude accounts in one click and says it works "even with live Claude sessions running" (Orca docs). Its Codex hot swap is also a choice you make: "any new Codex session launched after that uses it", while "sessions already running keep their original account until restarted" (Orca docs). Small command-line switchers exist too; we compared them.

If you'd rather swap by hand when a warning shows, those tools do it well. Where we differ is narrower: Shuriken swaps automatically, inside the running chat, on every device. That matters at 3 a.m. and during a meeting, when nobody is there to click.

Try it with two accounts#

Switching is on every plan, Free included, and Free holds two sources: enough to see your first hot swap. Add the accounts you already pay for, and the next time one reaches 99%, look for the card. Try Shuriken.

Sources

  1. Orca docs: Hot-swap Codex accounts (accessed 22 October 2026)
  2. Orca docs: Claude Code (accessed 22 October 2026)
  3. Claude Code docs: How Claude Code uses prompt caching (accessed 22 October 2026)

Frequently asked questions

What is hot swapping AI accounts?

It's our name for Shuriken moving your running chats to another of your accounts or sources when the one in use nears its limit. The chat keeps its history and carries on; you don't log out or start over.

When does a hot swap happen?

At whichever comes first: the account in use reaching your switch point (99% of a window by default; you can set anything from 50% to 100%), or the provider reporting a limit.

Does hot swapping give me more usage than I pay for?

No. It only uses the plans and keys you add yourself. Use accounts you own and are entitled to use; providers' terms may not allow sharing a subscription sign-in.

Can it spend money on an API key without asking?

Not a key from the chat's own provider: Switch to API keys is off until you turn it on, and even then a key takes chats only once every subscription is out, with a spend cap per fallback run of $25 unless you change it. An endpoint you added that serves another company's models can take a chat between turns once every source of the chat's own company is out.

Free tools