Joule Cloud docs

Compute

Container-shaped workloads on the mesh. Bring an OCI image, declare it in invisible.hcl, ship. The carbon-aware scheduler picks the cheapest cleanest grid that meets your region constraints; you get a live URL plus a continuous joule readout. No reserved instances. No idle billing.

The deploy file

workload "api" {
  image  = "ghcr.io/me/myapi:0.3.1"
  region = "auto"                # cheapest cleanest grid
  cpu    = "2 vCPU"              # min spec
  memory = "4 GB"
  energy_budget = "10 kJ/day"    # hard ceiling, optional
  scale {
    min = 1
    max = 20
    on_metric = "requests_per_sec > 100"
  }
}

route "api.example.com" {
  to = workload.api
}

What's different from EC2 / Compute Engine / Fly Machines

WhatHow we do it
BillingJoules consumed at the silicon, not vCPU-hours. An idle container near-zero. A pegged CPU at full power.
PlacementCarbon-aware scheduler picks region. Override with region: "eu-fi".
ScalingDeclarative in invisible.hcl. Scale-to-zero with one-shot warm starts.
NetworkZero inter-node-egress fees on the mesh. Public ingress free up to fair-use threshold.
IsolationFirecracker microVMs. No noisy-neighbour effects.
ReceiptsPer-request and per-hour joule receipts, signed.

Energy budget — the runaway-loop killer

The energy_budget field is a hard ceiling. If your workload exceeds the daily / hourly cap, the scheduler suspends it and emails you. Useful for batch jobs you don't want to discover at the end of the month after a stuck loop ate a year of compute.

CLI

invisible login
invisible deploy
invisible logs api
invisible energy api --since 1h
invisible scale api --min 3 --max 50
invisible suspend api
invisible delete api

Limits at launch