Rent a GPU With One API Call — Introducing the ParalonCloud Rental API

Starting today, you can rent a GPU on ParalonCloud without touching the browser. The new Rental API does everything the Rent a GPU page does — find a GPU, start it, connect, stop it — over a plain REST API. If your workflow lives in a script, a CI pipeline, or an AI agent, you no longer have to click through a UI to get compute.

And it uses the same prlc_ key you already use for the Inference API. One key, one balance, two APIs.

Rent a GPU in four calls

Renting is asynchronous — starting a GPU isn't instant, so the flow is find → start → poll → stop:

KEY="prlc_your_key_here"   # needs the 'rental' scope

# 1. Find a GPU (grab a node_id)
curl https://paraloncloud.com/api/v1/gpus -H "Authorization: Bearer $KEY"

# 2. Start it — returns a rental_id immediately, status "pending"
RID=$(curl -s -X POST https://paraloncloud.com/api/v1/rentals \
  -H "Authorization: Bearer $KEY" \
  -H "Idempotency-Key: my-run-001" \
  -H "Content-Type: application/json" \
  -d '{"node_id": "<node_id>", "type": "jupyter"}' | jq -r .rental_id)

# 3. Poll until it's running, then read the connection URL + token
until [ "$(curl -s https://paraloncloud.com/api/v1/rentals/$RID \
  -H "Authorization: Bearer $KEY" | jq -r .status)" = "running" ]; do sleep 3; done
curl -s https://paraloncloud.com/api/v1/rentals/$RID -H "Authorization: Bearer $KEY" | jq .connection

# 4. Stop it (stops billing)
curl -X DELETE https://paraloncloud.com/api/v1/rentals/$RID -H "Authorization: Bearer $KEY"

That's the whole thing. Full reference in Create & manage rentals.

Browse real inventory, with the numbers that matter

GET /gpus returns the nodes you can rent right now — online, priced, and free — with everything you need to choose:

• price_per_hour — the rate, locked for the life of the rental.
• GPU model and vram_mb — the card and its memory.
• compute_cap — the CUDA compute capability. This is the field people forget: a cheap node with lots of VRAM but a low compute capability may not run a modern stack at all. For FP8 LLM inference you want 8.9 or higher. We surface it so you can filter before you rent, not after.
• country / country_code — where the node is, for latency and data-residency.

You rent by node_id, a stable UUID — never by a name that two providers might share.

Built for money-safe automation

An API that spends credits needs guardrails. The Rental API has them by design:

• Scoped keys. A key can call inference out of the box, but it cannot start rentals until you grant it the rental scope in the Console. Every existing key stays inference-only. A key that leaks into a notebook or CI log can't rent GPUs on your dime unless you explicitly allowed it. See Authentication & scopes.
• Idempotent creates. Send an Idempotency-Key and a retried request returns the same rental — no accidental second GPU, no double billing.
• Auto-stop. Pass hours and the rental stops itself after that long, even if your script crashes and never calls DELETE. Open-ended rentals run until you stop them.
• Per-key limits. Each key has a max_active_rentals cap (set it in the Console) so a runaway loop can't spin up ten GPUs.
• A real stop. DELETE /rentals/{id} stops the GPU and billing, and it's safe to call twice.

Per-minute billing means you pay for exactly the minutes a rental runs — start it in code, tear it down in code, pay for the middle.

What this unlocks

• Agent-driven compute. An AI agent that needs a GPU for a task can rent one, use it, and release it — no human in the loop.
• CI/CD GPU jobs. Spin up a GPU for a test or a nightly training run, then destroy it.
• Batch pipelines. Fan out across GPUs on demand and clean up when the queue drains.
• On-demand fine-tuning. Provision, train, checkpoint, stop — from a single script.

Get a key and go

1. Open the Console and create an API key (or reuse your inference key).
2. Enable the GPU Rentals scope on it.
3. Point your script at https://paraloncloud.com/api/v1 and read the docs.

Prefer clicking first? The Console has a Rent tab with the same flow and a live view of your running rentals, and you can still rent from the Rent a GPU page. Check pricing to see what today's GPUs cost.

The Rental API and the Inference API share one key and one credit balance — build agents that both call models and rent the hardware to run them.