Rent seconds, not servers.

Computers for your code and AI agents. Ready in under a second, billed per second, asleep at $0 compute.

Scroll to watch it roll
Awake0:00:00
You pay$0.0000
Machine2 CPUs
How it works

Look inside the machine.

Tap a part
1

The API

Live

Everything you can click, your code and your agents can call.

POST /v1/machines
Idempotency-Key: 7c1e9a…
{"cpus": 4}

201 Created
{"id": "m_01k8f3x9…", "state": "running", "ready_ms": 740}
  • REST + JSON
  • OpenAPI 3.1
  • RFC 9457 errors
  • Idempotency keys
  • Rate limits
  • wm CLI
  • Python SDK
  • TypeScript SDK
  • MCP server
2

Compute

Live

Real processors and memory, fenced off. What you pay for is only yours.

$ cat /sys/fs/cgroup/cpu.max
400000 100000
$ cat /sys/fs/cgroup/memory.max
8589934592  # 8 GiB
  • cgroup v2 limits
  • 2 GB per vCPU
  • Own disk
  • Linux
  • Git · Python · Node · Go · Rust
  • Postgres · Chrome
3

Sleep and wake

Live

It naps when nothing reaches it and its programs go quiet. Never in the middle of a job.

$ wm machines create --idle-after 5m
pauses:  after 5m idle (nothing reached it, its programs quiet)
$ wm machines pause m_01k8f3x9…  # memory kept
  • Idle = no calls + CPU under 5%
  • cgroup cpu.stat
  • Freeze / thaw
  • Same PID after waking
  • 1 min to 24 h
  • --no-sleep
4

Files, copies, snapshots

Live

Files stay put. Copy a running machine in about a second, or keep it as a snapshot.

$ wm machines copy api-tests       # about 1 s
$ wm machines snapshot api-tests   # 4.5 s
$ wm machines create --from-snapshot api-tests
$ wm cp data.csv api-tests:work/data.csv
  • Persistent disk
  • Live copy
  • Snapshots
  • Start from a snapshot
  • Archive
  • wm cp
5

Where it runs

Live

Ask for a city. It runs near it, on a server with room, or on a new server made ready for it.

$ wm machines create --region mumbai
$ wm regions   # all 33: Mumbai, Delhi, Bengaluru,
               # Singapore, Tokyo, Frankfurt, London,
               # New York, São Paulo, Sydney…
  • 33 cities
  • --region
  • Region-aware placement
  • Capacity added on demand
6

A GPU on its back

Early access

Add a GPU when the work needs one. Opening account by account.

L40S$1.25 / h
A100$2.06 / h
H100$4.59 / h
H200$6.08 / h
$ wm gpus
$ wm machines create --gpu <id>
  • NVIDIA
  • Billed per second
  • Same API as CPU machines
7

Walls

Live

Your machine, your files, your keys. No way in for anyone else.

keys       kept as SHA-256 fingerprints
scopes     a key can never make a stronger key
inbound    no open ports
transport  TLS on every public connection
  • Own disk and processes
  • Hard CPU and memory limits
  • Scoped API keys
  • No inbound ports
  • Memory-safe Rust and Go

Everything in the box

LiveBuildingEarly accessNext
    Three commands

    Make it. Use it. Park it.

    wm · a terminal
    1. 1 Make
    2. 2 Run
    3. 3 Park
    
              
    A demo, printed the way wm prints
    Statenone
    Awake0:00:00
    You pay$0.0000
    Who rolls with us

    Made for people who build.

    Developers

    CPU

    Engineers

    CPU or GPU

    Researchers

    GPU

    Robotics

    GPU

    Data

    CPU

    AI agents

    CPU or GPU
    Price

    Pay by the second.

    CPU
    GPU
    25 min
    You pay
    $0.017
    By the hour $0.04
    Asleep: $0 compute Free downloads No minimum
    Compare

    How far $10 goes.

    Size
    CPU busy
    How we count

    Machine-hours that $10 buys at each company's own list price, read on 22 October 2026, for the size chosen with 2 GB of memory per CPU. Vercel, Cloudflare and Fly.io Sprites charge for the CPU only while it works; the others charge for the whole time a machine runs. Cloudflare needs at least 3 GB per CPU, so it is counted with 3 GB. Modal counted at 2 CPUs per core, its cheaper reading. Fly.io Machines with dedicated CPUs. Plan fees not counted. Wheelmachine disks: $0.10 per GB a month.

    Soon

    Your licence. Our wheels.

    Your licence server Private tunnel Allow-listed

    Set the world in motion.