Documentation

Build on Newcastle Compute.

Everything below assumes you have an API key. Get one. The API is at https://api.compute.newcastlerising.com.au/v1. Authentication is a bearer token in the Authorization header.

Authentication

Bearer tokens

Every request needs an Authorization: Bearer YOUR_KEY header. Keys are scoped per-organisation. Rotate keys from the dashboard; revoked keys stop working within seconds.

export NC_TOKEN="nc_live_..."

curl https://api.compute.newcastlerising.com.au/v1/usage \
  -H "Authorization: Bearer $NC_TOKEN"

Compute API

Provision and manage instances

Launch an instance

POST /v1/instances
Authorization: Bearer $NC_TOKEN
Content-Type: application/json

{
  "hardware":  "h200-141gb",
  "count":     1,
  "image":     "ubuntu-24.04-cuda-12",
  "ssh_keys":  ["ssh-ed25519 AAAA..."],
  "volumes":   [{"size_gb": 200, "mount": "/data"}]
}

List instances

GET /v1/instances

Stop and terminate

POST /v1/instances/:id/stop
DELETE /v1/instances/:id

Usage and billing

GET /v1/usage?from=2026-05-01&to=2026-05-31

Hardware identifiers: b200-192gb, h200-141gb, h100-80gb, rtx-pro-6000-blackwell, mi325x-256gb.

Inference API

OpenAI-compatible chat completions

Same shape as the OpenAI chat completions API. Most client libraries work by swapping the base URL and API key.

Chat completion

POST /v1/chat/completions
Authorization: Bearer $NC_TOKEN

{
  "model": "qwen-3-coder",
  "messages": [
    {"role": "user", "content": "Summarise this changelog in three sentences..."}
  ],
  "temperature": 0.3,
  "stream": false
}

Streaming

Pass "stream": true to receive Server-Sent Events. Tokens come back as standard data: {...} chunks ending with data: [DONE].

Available models

qwen-3-instruct — general-purpose, 128k context
qwen-3-coder — code generation and tool use
qwen-3-small — cheaper, faster, smaller

SDKs

Client libraries

Python

pip install newcastle-compute

Async-first, typed, drop-in compatible with the openai client for inference.

Go

go get newcastlerising.com.au/compute

Idiomatic Go, context-aware, no codegen step.

CLI

brew install newcastle-compute

Static binary for macOS, Linux, and Windows. nc command.

Limits and behaviour

Rate limits, retention, and quotas

Item	Default	Notes
API requests	60/sec per key	Raise on request; we'll usually agree.
Inference tokens	500k/min sustained	Burst higher; sustained needs reserved capacity.
Instance retention	Indefinite while running	Stopped instances kept for 7 days, then reaped.
Persistent volumes	Indefinite	Billed monthly at $0.12/GB. Deletable any time.
Egress (outside AU)	AUD $0.06/GB	Inside AU is free.

Troubleshooting

Common errors

`401 invalid_api_key`

Key was rotated or never issued. Generate a new one from the dashboard, or reach out if you can't.

`409 capacity_unavailable`

No instance of that hardware is currently free. The error includes an estimated_wait_seconds field. For predictable access, reserve capacity.

`429 rate_limited`

You've gone over the per-key rate limit. Back off, batch where you can, or raise the limit.

`500`

Email support with the request_id from the response. We aim to reply inside 30 minutes during business hours.

v1.0 documentation. The API is stabilising fast; we keep a public changelog and we will give you at least 30 days notice before any breaking change. See /blog for changelog and announcements.