Your first confidential call, in two minutes

One command for an inference call on a model running inside Intel TDX. Three for a full Confidential VM with an NVIDIA GPU, where you generate the attestation yourself. Billed per second from your balance. Prefer clicking? The same in the dashboard, in four steps.

Before you start

Create an account and add credit to your balance.
Create a key at voltagegpu.com/api-keys. There are two kinds: an inference key (starts with vgpu_, sent as a Bearer token) and a pods key (starts with volt_, sent in an X-API-Key header).
For a machine only: add your SSH public key at voltagegpu.com/ssh-keys first. A Confidential VM binds its key when it is created and cannot be given one afterwards.

PATH A · 30 SECONDS

An inference call

1One command

curl https://api.voltagegpu.com/v1/chat/completions \
  -H "Authorization: Bearer $VOLTAGE_INFERENCE_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "Qwen/Qwen3-32B-TEE", "messages": [{"role": "user", "content": "Hello"}]}'

2Or with the OpenAI SDK, unchanged

from openai import OpenAI

client = OpenAI(base_url="https://api.voltagegpu.com/v1", api_key="vgpu_...")
reply = client.chat.completions.create(
    model="Qwen/Qwen3-32B-TEE",
    messages=[{"role": "user", "content": "Hello"}],
)
print(reply.choices[0].message.content)

Use the exact model id: Qwen/Qwen3-32B-TEE, not a short name. It costs $0.19 per million input tokens and $0.77 per million output tokens right now. Every model is listed by GET /v1/models and on the pricing page. There is no confidential embedding model yet.

PATH B · ABOUT THREE MINUTES

A Confidential VM with a GPU

1Deploy

curl -X POST https://api.voltagegpu.com/api/volt/pods \
  -H "X-API-Key: $VOLTAGE_PODS_KEY" \
  -H "Content-Type: application/json" \
  -d '{"provider": "confidential", "name": "first-vm", "resource_name": "rtx6000b-small"}'

rtx6000b-small is one RTX PRO 6000 Blackwell (96 GB) in an Intel TDX trust domain, $3.80 per hour. One hour is charged upfront; stopping early gives back the unused part. Ubuntu 24.04 with the NVIDIA driver and CUDA preinstalled.

2Connect

curl https://api.voltagegpu.com/api/volt/pods -H "X-API-Key: $VOLTAGE_PODS_KEY"
# when ssh_ready is true, run the ssh_command it returns, for example:
ssh -p SSH_PORT ubuntu@SSH_HOST

The VM is usually ready in about two minutes.

3Release it

curl -X POST https://api.voltagegpu.com/api/volt/pods/POD_ID/stop \
  -H "X-API-Key: $VOLTAGE_PODS_KEY"

Stopping a Confidential VM destroys it and erases its disk, so copy out what you want to keep first.

Check it yourself

Inside the VM you generate an Intel TDX quote and the NVIDIA GPU report on a challenge you pick, and verify them on your own machine: the guide. For inference, this returns the evidence of the instances serving the model, bound to your nonce. It covers the instances, not each response:

curl "https://api.voltagegpu.com/v1/attestation?model=Qwen/Qwen3-32B-TEE&nonce=$(openssl rand -hex 32)" \
  -H "Authorization: Bearer $VOLTAGE_INFERENCE_KEY"

We also rent a VM every week and publish the result, raw files included: voltagegpu.com/proofs.

Next: the full getting-started guide, the API reference, or the OpenAI SDK page. Stuck? Write to contact@voltagegpu.com, a person answers.