Blog
The big updates, the stories behind them and what is planned next.
-
Shuriken is officially launched, in beta
The harness for Claude Code, Codex and other coding agents is open to everyone. It began as our internal tool; here is what it does and what beta means. -
How Shuriken started: a tool we built for ourselves
We build software for clients and ourselves. No harness fit how heavily we use AI, so we built our own. Then a programmer asked: why don't we sell this? -
We were tired of logging in and out of Claude
Limit reached, /logout, /login, paste the context back. We did it so often we built Shuriken: chats move to an account with room by themselves. -
The 5-hour clock starts when you do. So we start it.
Claude's 5-hour window only starts when you use an account, so an idle one wastes time. Shuriken starts its clock with one tiny background request. -
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%. -
How Shuriken picks the next account
At your switch point (99% by default) or a limit, Shuriken moves your chats to the source with the most room. Expiring capacity goes first, API keys last. -
Best MCP servers for Claude Code (2026)
Our MCP picks for Claude Code, by job: GitHub and Sentry, Linear and Notion, Context7 and DeepWiki, Playwright, Supabase, Neon, PostHog and Stripe. -
Claude Code and Codex in one project
How to use Claude Code and Codex together on one project: a chat per agent, one set of MCP servers for both, and each agent on its own sources. -
A git workflow for Claude Code that keeps main clean
The Claude Code git workflow we'd give anyone running several agents: a branch per task, small commits, read the diff, checks before merge, then squash. -
How to review AI-generated code without reading every line
We review agent diffs tests first: would each test fail without the change? Then deletions, the diff's edges, secrets, and a run. Our real misses. -
Test-driven development with Claude Code
Claude stops when work looks done, so give it a failing test to run. Our TDD prompt in three steps, and a Stop hook that runs tests when Claude finishes. -
Why every chat gets its own git worktree
Two coding agents in one folder break each other's builds. In Shuriken every chat but Main gets a git worktree: its own checkout and branch. -
Yes, it's called Cocaine mode
Cocaine mode keeps your Mac awake with the lid closed, so your coding agents keep working. Why it has that name, how it works, and the honest warnings. -
Every plan and key you own, working as one
Add your Claude, ChatGPT, Gemini and other plans and API keys to Shuriken as sources. When one runs out, your chats move to the next. -
Two Notion workspaces, one MCP list
Add Notion twice, once for work and once for personal, each with its own sign-in, and pick which projects get which. Switching accounts never asks again. -
Your agents run on your machines, not ours
Shuriken runs coding agents on your own computers and servers. Our cloud links devices and relays sealed messages. Why we chose local, and the trade-off. -
Bring your own plans: we never bill model usage
Shuriken runs Claude Code, Codex and other agents with your own subscriptions and keys. We never resell tokens or add a margin. -
Why chats said "Connection lost", and the fix
On 5.0.0, one very long chat could freeze Shuriken's background service: "not reached" blinking, window reloads, stopped agents. The cause and the fix.
The first posts go up on launch day.
Try Shuriken on your own computerYou're signed in
Add your Claude, Codex and Gemini accounts once. Shuriken switches between them so your agents keep working.Your accounts, projects and chats are in the app.
Coming soon