Shuriken is officially launched, in beta

Shuriken is officially launched. From today anyone can download it, make an account and run their coding agents with it. It's a harness for Claude Code, Codex and the other agents you already use: it runs them on your own computers and servers, moves your chats to another of your accounts when one hits its limit, and lets you follow everything from your desk or your phone.

We're calling it a beta, and we mean it.

From internal tool to launch#

Shuriken started as our internal tool. We build software for clients and for ourselves, and none of the harnesses we tried fit how heavily we use coding agents, so we built our own.

It didn't stay with the people who built it. It's used across our company's departments and in the other companies in our group, and we ran it almost like a product: when a provider changed something, we fixed it, and when something annoyed us twice, we changed it.

That made opening it up feel natural. When one of our programmers asked why we didn't sell it, it clicked right away: it was too useful to keep to ourselves. The whole story is in How Shuriken started.

You'll also find a lot to read from the first day. We wrote these guides, comparisons and notes while building Shuriken: how the Claude and Codex limits work, how the tools compare, and why we built each part the way we did. Today we publish them together, in the guides, the comparisons and on the blog, and new ones follow every day.

Three weeks of closed testing#

Shuriken didn't start today. The release pipeline built the first prerelease, v0.1.0-rc.1, for every platform on 29 September 2026. Since then we've run it on our own projects every day, together with a small group of early accounts.

1.0 came out on 4 October. By 19 October we were at 5.0.0, after 23 tagged releases in 21 days. The version number ran from 0 to 5 in three weeks because we shipped small releases often, sometimes three in a day.

Closed testing did its job: it found things. The worst was a single very long chat that could freeze the background service on 5.0.0 and make a computer look offline. We found it on our own machine on 20 October, and the fix is in 5.1. The full story is in Why chats said "Connection lost", and the fix.

We open with Shuriken 5.1. 5.1.1 came out on 25 October with sign-in through Google, Apple or an emailed code, and the update we ship just before the launch adds running a chat on another computer, Planning mode and a new look: a white accent and a new logo. What's new in Shuriken 5.1 has the rest.

Why it exists#

The short version: we kept hitting Claude's limits in the middle of a task, logging out, logging in to another of our own accounts, and pasting the context back in. We wanted the chat to just keep going on an account that still had room.

The long version, with more complaining, is in We were tired of logging in and out of Claude.

What it does#

• Automatic source switching. Add the Claude accounts, ChatGPT plans and API keys you own and are entitled to use. We call each one a source. When the one in use reaches your switch point (99% by default), your chats move to the one with the most room, on every device at once. How switching works.
• 5-hour clocks. Claude's 5-hour window starts when an account is first used. Shuriken starts idle accounts' clocks with one tiny background request each, so your limits come back as early as they can. 5-hour clocks.
• Global MCPs. Sign in to an MCP server once and it works in every project you give it, on any account and any device. Global MCPs.
• 67 coding agents, in parallel. Claude Code, Codex, Gemini CLI and the rest, each chat on its own branch in its own git worktree, with no cap from us. Parallel agents.
• Run a chat on another computer. Pick Run on… in a chat and choose another of your computers or servers. The chat moves with its history, settings and uncommitted work, stays where it was in your list, and carries on there. How it works.
• Planning. Pick Planning in the message box and the agent plans first, in its own plan mode, and asks before it changes anything.
• Cocaine mode. Yes, really. Pick which computers stay awake; on a Mac they keep working with the lid closed. Why it's called that.
• Design mode in the browser. Open your running app inside Shuriken, click an element, type or say what should change, and the agent gets the element's details and a cropped screenshot. The browser.

Where it runs#

• Desktop: macOS 13 or later on Apple silicon (M1 or newer), Windows 10 and 11, and Linux (AppImage and .deb).
• Servers: add a Linux, macOS or Windows machine as a headless server and run agents there. Servers.
• Your phone: the web app, in any phone browser. It shows your chats, the queue and changes, and you can install it to your home screen. Native iOS and Android apps are coming in November. Phone.

Agents, code and chat history stay on your machines. Our cloud signs you in, links your devices and relays end-to-end encrypted messages between them. It runs no agents and stores no code. More on that in Your agents run on your machines, not ours.

And one thing we'll keep repeating: Shuriken never bills model usage. Your agents run on your own plans and keys. Every new account starts with 14 days of Pro, free and with no card. After that it moves to Free unless you upgrade, and Free is a plan, not a trial: two sources and two linked devices. Pro is €49 a month, or €490 a year. If you signed up before today, you're a Founding member, with Pro's features at no charge for as long as Shuriken runs.

What "beta" means here#

Small releases, often. Some things will break. When they do, press the bug button in the top bar: it attaches screenshots of the window, and "What's new" tells you when your bug is fixed. Or write to office@shuriken.sh.

Two honest gaps while we're at it. The Mac app is signed and notarised by Apple, but the Windows installer isn't code-signed yet, so SmartScreen may warn you once before it runs. And on your phone, Shuriken is the web app until the native apps land.

If you run more than one coding agent, or more than one account, we built this for you. Try Shuriken and tell us what breaks.