Every plan and key you own, working as one

Most people we talk to have more than one AI plan. A personal Claude subscription and a work one. A ChatGPT plan for Codex. A Gemini key from a side project. An OpenRouter key nobody remembers making.

Each of those lives in its own app, its own terminal, its own login. When one runs out at 3 pm, you log out, log in to another, re-open the project and explain everything to the agent again. We did this for months. Then we built Shuriken so we would never do it again, and the idea at its centre is the source.

What a source is#

A source is a subscription sign-in or an API key you add to Shuriken. That's all.

• Your Claude Pro or Max plan, signed in: a source.
• Your ChatGPT plan, which Codex runs on: a source.
• A Gemini, Copilot, Cursor, Grok, Kimi or Qwen sign-in or key: a source each.
• Your own endpoint: a source too.

You add them once, under Sources, and Shuriken keeps them in its encrypted vault, synced between your devices. The cloud never sees the sign-ins or keys themselves.

Several per provider, one in use#

You can add as many sources per provider as you have. Three Claude accounts? Add all three.

At any moment, one source per provider is in use, and it is the same one on every computer and server you have linked. Press Use on your laptop, and your server follows. No tab per provider, no "which login is this terminal on?"

Why it matters: chats move by themselves#

Here is the part we built Shuriken for. When the source in use reaches your switch point (99% by default, and you can set it anywhere from 50% to 100%), or the provider reports a limit, your chats move to the source of that provider with the most room left. On every device at once.

A Claude Code chat moves on its very next request, subagents and all. Other agents move when their current turn ends. Either way, a small "Switched to …" card in the chat tells you what happened. You don't log out, you don't log in, and you don't hand the project to anyone.

Three rules keep this sane.

• Your provider's sources come first. A Claude Code chat moves to another of your Claude sources, a Codex chat to another ChatGPT sign-in. Only when every one of them is out can the chat go on, between turns, to another company's endpoint you added for that same agent (say, a Kimi endpoint behind Claude Code).
• API keys go last, and only if you say so. With "Switch to API keys" on (it starts off), a key takes chats only once every subscription is out. A spend cap per fallback run, $25 by default, counts what the run spends across every key it uses and stops it there.
• Broken sources are skipped. A free plan, a cancelled subscription or a signed-out account never gets a chat switched onto it.

By default a chat stays on one source until it reaches your switch point, then moves to the next (How: When one runs out). You can pick Spread the work instead, which moves chats to the source with the most left on a clock, every 15, 20, 30, 45 or 60 minutes. These settings are under Switch settings at the top of Sources. The source switching page covers every setting.

Which agents can use them#

Sources feed agents. Shuriken runs 67 coding agents, Claude Code, Codex and Gemini CLI among them, and each one uses the sign-ins you already have. How deep Shuriken goes varies by agent. Claude Code gets the most: switching mid-task, subagents in Activity, and so on.

If you mostly live in Codex, our post on running Codex with several ChatGPT accounts goes into the details.

Free holds two, Pro holds the rest#

On the Free plan you can add 2 sources. That is enough for one personal and one work Claude plan, or a Claude plan and a ChatGPT plan. Free is a plan, not a trial, and it needs no card. Automatic switching and 5-hour clocks are included on Free.

Pro removes the limit: add every source you own. The pricing page has the current details.

Either way, Shuriken never bills model usage. Your agents run on your plans and your keys, and your provider bills you as it always has.

A word on terms#

Sources are for plans and keys you own and are entitled to use. Switching between your own accounts is what Shuriken does. It is not a way around a provider's rules: don't buy or borrow accounts to stretch limits, and check your providers' terms. Some don't allow sharing a subscription sign-in with another person, and doing so can put the account at risk. Where you want to give someone access, an API key is the safer route.

Shuriken 5.1 added local models as sources. Point a source at Ollama, LM Studio or any Anthropic- or OpenAI-compatible address. Its turns aren't priced; they read "local". It's an endpoint, so it never goes ahead of a subscription: it takes chats when you press Use on it, or once every subscription is out (by itself for Claude Code and Codex, with "Switch to API keys" on for the rest). It also added tags (Work, Personal, Client) to filter the Sources list.

If you have more plans than tabs, try Shuriken, add them once, and let your chats find the room.