Developer portal
One API and one CLI for the entire GPU fleet
Deploy, inspect and terminate H100, A100 and B200 clusters over REST — or run the exact same operation with kwctl. Full parity, per-second billing, $0 egress.
$ kwctl auth login$ kwctl keys create --name enterprise-prod$ kwctl deploy --gpu h100 --count 8Portal capabilities
Keys, analytics and quotas in one place
API Key Generation
Mint scoped kw_live_ keys per environment or tenant. Shown once, hashed at rest, revocable instantly from the console or the tokens endpoint.
Usage Analytics
Per-key request volume, GPU-hours consumed, p99 latency and error rate — streamed to the console and exportable to your own observability stack.
Quota Management
Rate limits, concurrent-cluster ceilings and spend caps enforced at the edge, with deterministic 429s and Retry-After headers.
30
Requests / minute
Per API key · 429 when exceeded
3
Concurrent jobs
Per API key · raised on request
Customer set
Monthly spend cap
Optional ceiling above balance
Up to 365d
Key TTL
Instant revocation, hashed at rest
REST API reference
Orchestration endpoints
https://api.kilawattcloud.dev/v1/compute/deployProvisions a bare-metal GPU cluster (H100, A100, B200) on the requested fabric and region. Returns once nodes pass pre-flight health gating — typically under 45 seconds.
Parameters
| Field | Type | Required | Description |
|---|---|---|---|
| gpu | string | required | nvidia-h100 · nvidia-a100 · nvidia-b200 |
| count | integer | required | GPU count, 1–512 |
| region | string | required | us-east · us-west · eu-central |
| image | string | optional | Container image (default: pytorch-2.3) |
| tier | string | optional | on-demand · reserved-1y · dedicated-3y |
{
"gpu": "nvidia-h100",
"count": 8,
"region": "us-east",
"image": "pytorch-2.3",
"tier": "on-demand"
}{
"id": "clu_8fj2k1x9",
"status": "provisioning",
"gpu": "nvidia-h100",
"count": 8,
"region": "us-east",
"fabric": "quantum-2-ndr-infiniband",
"ssh": "ubuntu@clu-8fj2k1x9.kilawattcloud.dev",
"created_at": "2026-08-27T14:02:11Z"
}Integrate
curl -X POST https://api.kilawattcloud.dev/v1/compute/deploy \
-H "Authorization: Bearer $KILAWATT_API_KEY" \
-H "Content-Type: application/json" \
-d '{"gpu":"nvidia-h100","count":8,"region":"us-east"}'Billing, limits & safety
How spend is authorized and enforced
Full-cost pre-authorization
Before any node is contacted we price the whole job — GPU count × rate × requested duration — and reject it with a 402 naming the exact amount required if your balance can't cover it. No partial-hour launches.
Per-key rate & concurrency limits
Each key is capped at 3 concurrent active jobs and 30 requests per minute by default, returning 429 when exceeded. Limits are enforced server-side and can only be raised by Kilawatt.
Customer-set monthly spend cap
Set an optional monthly ceiling in Billing & Usage, per account or per key. Jobs that would push the month's usage past the cap are refused with a 402 even when the wallet is funded.
Live metering & automatic shutdown
Every instance we start — console or API, on either compute node — is tracked with a paid-through deadline. A job meters elapsed time every 15 minutes and terminates the instance at its node when the balance reaches zero or the paid window expires.
401
Invalid or revoked key
Authorization: Bearer kw_live_… missing, unknown or inactive
402
Insufficient balance or cap hit
Body names the exact USD required for the full job
429
Rate or concurrency limit
30 req/min or 3 concurrent jobs per key
503
No node available
All eligible compute nodes failed over; safe to retry
Quick integration
Connect in under a minute
import os, requests
r = requests.post(
"https://api.kilawattcloud.dev/v1/compute/deploy",
headers={"Authorization": f"Bearer {os.environ['KILAWATT_API_KEY']}"},
json={"gpu": "nvidia-h100", "count": 8, "region": "us-east"},
)
print(r.json()["id"], r.json()["status"])curl -X POST https://api.kilawattcloud.dev/v1/compute/deploy \
-H "Authorization: Bearer $KILAWATT_API_KEY" \
-d '{"gpu":"nvidia-h100","count":8,"region":"us-east"}'import Kilawatt from "kilawatt-js";
const client = new Kilawatt({ apiKey: process.env.KILAWATT_API_KEY });
const cluster = await client.compute.deploy({
gpu: "nvidia-h100", count: 8, region: "us-east",
});npm i kilawatt-jscurl -fsSL https://get.kilawattcloud.dev | sh && kwctl auth loginNote — kilawatt-js is the only first-party SDK. A dedicated Python SDK is coming soon; use the REST API directly (shown above) in the meantime.
Enterprise integration
Built for procurement-grade deployments
Full-cost pre-authorization
Jobs are paid in full up front — the API returns a 402 with the exact amount before any node is contacted.
Enforced spend safety
Per-key rate limits, concurrency ceilings and optional monthly spend caps, all enforced server-side.
Ephemeral workloads
Instances are metered every 15 minutes and automatically terminated at the provider when balance or the paid window ends.
$0 data egress
Move datasets and checkpoints out without egress fees on any plan.
Company registration
Los Angeles County Fictitious Business Name statement
Official public filing for Kilawatt Cloud, a registered fictitious business name of Right Recruit LLC. Sensitive addresses, signatures, and the filing barcode are redacted in the public copy.
