Joule Cloud documentation

Joule Cloud is a green neocloud. You ship a workload, the router picks the cheapest capable silicon across our mesh, and you're billed in joules — the unit physics actually charges for computation.

The same APIs you already use, lower bill, and a receipt for every workload.

Choose your path

• I want to run my first request in 60 seconds. → Quickstart
• I want the full endpoint list and request shapes. → API reference
• I want to know what it'll cost. → Pricing
• I want to understand what "billed in joules" means. → What is a joule, here

What is a joule, here

A joule is one watt-second — the SI unit of energy. Joule Cloud measures the energy each request actually consumes at the silicon, using hardware counters: Intel RAPL on x86, NVIDIA NVML on GPUs, IOReport on Apple Silicon. The number is captured by the gateway and returned to you in the X-Energy-Joules response header of every request, and totalled on your bill.

You pay for that number times the per-joule rate. Idle costs near nothing. Inefficient code shows up. Efficient code shows up.

Routing & placement

For inference requests, the router classifies each request into a cost tier — lookup, extraction, aggregation, reasoning — and routes it to the cheapest capable silicon currently available across the mesh. For deployed workloads, you can pin a region, or set region = "auto" in your invisible.hcl and let the carbon-aware scheduler place it where the grid is cleanest.

Energy receipts

Every request and every workload run produces a receipt: the workload identifier, the silicon that executed it, the energy in joules, the local grid carbon intensity, the operator's published PUE, and a cryptographic signature. Receipts are visible in the console and can be exported in bulk for audit.

Where to go next

If you're building, start with the Quickstart. If you're evaluating, read the Pricing page. If you're integrating an existing application, the API reference mirrors the OpenAI and S3 shapes you already know.