Snapshots Are A Crutch. Boot True-Cold In An Instant.
A fresh Linux kernel in 59ms. Six parallel Chibi Sub-Machines in 2ms each. On-demand workspaces when you need state, nothing left behind when you don't.
刹那 setsuna — n. an instant; the shortest moment of time
Request to ready
The race in numbers. Measured spans differ; see the full ColdBench report for methodology and sources.
Meet Machines & Chibis.
What The Hell Is A Chibi?
ちび chibi — n. a small child; a miniature thing
Setsuna provides a unique nested ecosystem that's optimized for parallel workload execution. Spin up a Machine and get a true-cold hardware boundary in milliseconds; then subdivide that Machine into Chibis for instant 2–3ms Sub-Machines. Customize and scale it however you want.
Your hardware-isolated foundation. A true-cold microVM with its own dedicated Linux kernel booted from scratch—never resumed from a stale snapshot. Pick your vCPU and memory tier, and it becomes the blast-radius boundary for everything inside it.
Isolated Sub-Machines. They spawn inside your running Machine in 2–3ms, each with its own processes, filesystem view, and resource limits. Slice your Machine’s CPU and RAM into 2, 6, or 16+ Chibis on the fly; they share the Machine’s kernel inside its microVM boundary at zero extra cost.
Built For True-Cold, Not Snapshots
Most sandbox platforms unpause cloned memory snapshots to mask multi-second cold starts. Setsuna’s custom runtime boots a genuinely fresh kernel from zero in milliseconds—no recycled memory state, no warm-pool drift.
Contained At The Parent Boundary
Every Chibi runs inside the dedicated kernel of its Parent Machine—never the underlying host. If untrusted or hallucinated code crashes the kernel, the blast radius stops at your disposable Machine while the rest of your fleet runs untouched.
Parallelism Without The Multiplier
Other clouds charge you a new sandbox bill every time a workload fans out. With Setsuna, you pay the per-second rate of your Parent Machine and slice it into as many Chibis as you need at zero additional cost.
Keep Compute Stateless.
Mount Storage On Demand.
Dragging gigabytes of files into every boot slows everything down. Setsuna launches every Machine completely state-free, then lets you mount persistent Workspaces straight from the SDK—attaching your repos, caches, and folders to a Parent Machine or individual Chibis whenever you need them.
01 · Boot Stateless — Machine & Chibis spin up completely empty—zero files dragged into boot.
> const machine = await setsuna.machines.create({ machineConfigurationId: '2vcpu-16gib' });
✔ Machine ready (clean state-free rootfs)
> await machine.workspaces.mount({ workspaceId: 'ws_repo', path: '/mnt/repo' });
✔ Mounted /mnt/repo, /mnt/cache, /mnt/out on demand
✔ Job finished — Machine destroyed, Workspace preserved.
Modular By Design
Keep your whole setup in one workspace or split GitHub repos, caches, and agent outputs into separate volumes that you mix and match across Machines and Chibis.
Outlives Every Machine
When a Machine and its Chibis vanish on exit, your Workspace remains intact for the next instant.
Workspace sizing & billing →Curated Compute That Matches Your Workflow
Two ways to buy: pay per second with Metered, or hold monthly capacity with Reserved Machines. Both run on Setsuna's production capacity with automatic placement.