Serving glm.gpu.justratel.com

EU-hosted GPU inference

Your own GPU in the EU. Your own endpoint. Billed by the hour.

Reserve an EU GPU, deploy an open-weight model, and get an OpenAI-compatible endpoint. Never shared, never metered per token.

  • EU-based
  • GDPR-aligned
  • Flat-rate

How it works

  1. Step 1

    Reserve EU resources

    Pick a GPU tier and an EU region. The card is yours alone; we never co-locate another tenant on it.

  2. Step 2

    Deploy a template

    Choose an open-weight model. Its weights are already baked into a disk snapshot, so the box boots serving instead of downloading.

  3. Step 3

    Call your endpoint

    Every deployment gets its own hostname and its own API key, minted onto the box. Point any OpenAI client at it.

Your private endpoint
from openai import OpenAI

client = OpenAI(
    base_url="https://qwen.gpu.justratel.com/v1",
    api_key="YOUR_API_KEY",
)

OpenAI-compatible from the first request. Change the base URL; change nothing else.

Why Ratel?

  • EU data residency

    Your GPU is pinned to an EU region, and your traffic goes straight to it. The control plane is not in the data path, so we cannot read a prompt.

  • Dedicated GPUs you own

    Root SSH on the box, snapshots, and one-click revert. Nothing else runs on the card.

  • One-click templates

    Each build ships with its serving arguments already chosen — quantization, context length, tensor parallelism. You pick the model, not the flags.

  • Scheduled start/stop

    Auto-start at 6AM, auto-stop at 8PM. Set a monthly budget cap and we stop provisioning when you reach it.

  • OpenAI-compatible endpoint

    Standard /v1/chat/completions on your own hostname. Any OpenAI SDK, in any language, with no Ratel library to install.

  • True flat-rate pricing

    You pay for the hours the card is powered on. No per-token fees, no rate limits, and no charge for what you run on it.

Ready to deploy?

Anyone can sign up. Our team turns on GPU access for each account, so tell us what you need to run.