API quickstart

This walks the whole path once: mint a key, find capacity, price it, rent it, connect, stop it. Every response below is real, trimmed only for length.

Mint a key

In the dashboard, go to Settings → API keys → Create key. Give it a name you will recognise in six months, tick the scopes it needs — for this walkthrough, all three — and copy the key. It is shown once.

export GPUOUTLET_API_KEY="gpk_live_…"

Confirm it works:

curl -s https://api.gpuoutlet.ai/v1/me \
  -H "Authorization: Bearer $GPUOUTLET_API_KEY"

{ "key_id": "key_7f3a2b1c", "scopes": ["catalog:read", "rentals:read", "rentals:write"], "mode": "live" }

Find capacity

Ask for H100s that are free right now, cheapest first:

curl -s -G https://api.gpuoutlet.ai/v1/offers \
  -H "Authorization: Bearer $GPUOUTLET_API_KEY" \
  -d gpu_family=h100-80gb \
  -d available=true \
  -d sort=price_asc \
  -d limit=3

{
  "data": [
    {
      "id": "off_3kd91mz7",
      "gpu_model": "h100-sxm-80gb",
      "gpu_name": "NVIDIA H100 SXM",
      "gpu_family": "h100-80gb",
      "vram_gb": 80,
      "max_gpus": 8,
      "instance_class": "standard",
      "region": "us-east-1",
      "region_name": "US East",
      "currency": "usd",
      "price_per_hour_cents": 249,
      "availability": { "kind": "count", "count": 6 },
      "service_tiers": ["on_demand", "spot"]
    }
  ],
  "meta": { "next_cursor": "eyJvIjoiMjQ5Iiwi…" }
}

Every filter is documented in Browsing the catalog.

Price it

A quote tells you what the rental costs per hour at the size you want, and whether it can start right now.

curl -s https://api.gpuoutlet.ai/v1/rental-quotes \
  -H "Authorization: Bearer $GPUOUTLET_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "offer_id": "off_3kd91mz7", "gpu_count": 2, "hours": 6 }'

{
  "offer_id": "off_3kd91mz7",
  "gpu_count": 2,
  "service_tier": "on_demand",
  "currency": "usd",
  "price_per_hour_cents": 498,
  "hold_cents": 498,
  "estimated_total_cents": 2988,
  "available": true,
  "sufficient_balance": true
}

price_per_hour_cents is for the whole rental, not per GPU. A quote reserves nothing — see Quotes.

Rent it

Creating a rental spends money, so Idempotency-Key is required. Generate one per intent and reuse it verbatim on every retry of that same intent.

curl -s https://api.gpuoutlet.ai/v1/rentals \
  -H "Authorization: Bearer $GPUOUTLET_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{ "offer_id": "off_3kd91mz7", "gpu_count": 2, "name": "sweep-run-14" }'

{
  "id": "rnt_8h2k4m",
  "status": "provisioning",
  "name": "sweep-run-14",
  "gpu_count": 2,
  "billing": { "currency": "usd", "price_per_hour_cents": 498, "accrued_cents": 0, "hold_cents": 498 },
  "access": null,
  "created_at": "2026-07-29T14:02:11.000Z"
}

provisioning means the machine does not exist yet. access is null and nothing is reachable. Do not treat a 201 as “ready”.

Wait for it, then connect

Poll the rental until it is running — usually well under a minute.

curl -s https://api.gpuoutlet.ai/v1/rentals/rnt_8h2k4m \
  -H "Authorization: Bearer $GPUOUTLET_API_KEY"

{
  "id": "rnt_8h2k4m",
  "status": "running",
  "access": {
    "ssh": { "host": "ssh.us-east-1.gpuoutlet.ai", "port": 20041, "username": "root" },
    "jupyter": { "url": "https://rnt-8h2k4m.gpuoutlet.ai/lab" },
    "web_terminal": null,
    "service": null
  },
  "billing": { "currency": "usd", "price_per_hour_cents": 498, "accrued_cents": 12, "hold_cents": 498 },
  "started_at": "2026-07-29T14:02:47.000Z"
}

ssh -p 20041 root@ssh.us-east-1.gpuoutlet.ai

Poll about once every two to three seconds. Every member of access is independently nullable — an offer may expose SSH but no Jupyter — so check before you use one.

Stop it

curl -s -X POST https://api.gpuoutlet.ai/v1/rentals/rnt_8h2k4m/stop \
  -H "Authorization: Bearer $GPUOUTLET_API_KEY"

{ "id": "rnt_8h2k4m", "status": "stopping", "billing": { "accrued_cents": 1743, "hold_cents": 498 } }

stopping is not stopped. Teardown is asynchronous and the final charge lands when the machine is actually gone. Poll until stopped before you trust accrued_cents as final, or before starting a replacement if you are near the concurrency ceiling.

The same thing in one script

import os, time, uuid, requests

API = "https://api.gpuoutlet.ai/v1"
S = requests.Session()
S.headers["Authorization"] = f"Bearer {os.environ['GPUOUTLET_API_KEY']}"

offers = S.get(f"{API}/offers", params={
    "gpu_family": "h100-80gb", "available": "true", "sort": "price_asc", "limit": 1
}).json()["data"]
if not offers:
    raise SystemExit("no H100 capacity right now")

rental = S.post(f"{API}/rentals",
    headers={"Idempotency-Key": str(uuid.uuid4())},
    json={"offer_id": offers[0]["id"], "gpu_count": 1, "name": "quickstart"},
).json()

# Poll until it is usable — or until it gives up. Both are terminal for us.
while rental["status"] == "provisioning":
    time.sleep(3)
    rental = S.get(f"{API}/rentals/{rental['id']}").json()

if rental["status"] == "failed":
    raise SystemExit(f"failed: {rental['failure_code']}")

print("ssh:", rental["access"]["ssh"])
S.post(f"{API}/rentals/{rental['id']}/stop")

const API = 'https://api.gpuoutlet.ai/v1';
const auth = { Authorization: `Bearer ${process.env.GPUOUTLET_API_KEY}` };
const get = (p) => fetch(`${API}${p}`, { headers: auth }).then((r) => r.json());

const { data: offers } = await get(
  '/offers?gpu_family=h100-80gb&available=true&sort=price_asc&limit=1',
);
if (!offers.length) throw new Error('no H100 capacity right now');

let rental = await fetch(`${API}/rentals`, {
  method: 'POST',
  headers: { ...auth, 'Content-Type': 'application/json', 'Idempotency-Key': crypto.randomUUID() },
  body: JSON.stringify({ offer_id: offers[0].id, gpu_count: 1, name: 'quickstart' }),
}).then((r) => r.json());

while (rental.status === 'provisioning') {
  await new Promise((r) => setTimeout(r, 3000));
  rental = await get(`/rentals/${rental.id}`);
}
if (rental.status === 'failed') throw new Error(`failed: ${rental.failure_code}`);

console.log('ssh:', rental.access.ssh);
await fetch(`${API}/rentals/${rental.id}/stop`, { method: 'POST', headers: auth });

Both scripts loop only while the status is provisioning, and treat everything else as an exit condition. Looping on !== 'running' instead would spin forever on a failed rental — the single most common bug in a first integration.

Get started

API keys

Conventions

Catalog

Rentals

Radar

The same thing in one script

Next

Every catalog filter

Rental lifecycle in detail

Spend limits and safety rails

Error reference

​The same thing in one script

​Next

Every catalog filter

Rental lifecycle in detail

Spend limits and safety rails

Error reference

The same thing in one script

Next