What if no GPU is free?
Before you book, the page shows a recent live check of free GPUs for your exact choice. Join the waitlist — we deploy as soon as a GPU is free. Nothing is billed while you wait. We keep trying for a limited time and show the deadline when you join. You can cancel any time. A check is an observation, not a reservation, and we can’t promise a GPU will free up. Where the model is also available as a hosted API we point you to it.
What exactly do I get?
A private, OpenAI-compatible endpoint that serves only MetaDecision-31B-Rain on GPUs reserved for you, with API keys you control, a live status page, request counts and spend. The model is pinned to an exact published revision.
How is it billed?
Your prepaid Decision Models balance is wallet credit, the same as for the API; it may need a top-up before we can start, and the top-up itself is not a hosting charge. Hosting billing starts only when your endpoint is ready: nothing is charged while a request waits for a free GPU or while the model downloads and loads. After that it is billed from your balance, in blocks: one hour at a time for “always on”, 15 minutes at a time while warm for “scale to zero”. VAT is applied when you top up. The next block is charged just before the current one ends. When your balance can’t cover the next block, the endpoint stops — you are never billed beyond your balance. Optional daily and monthly caps.
What does “scale to zero” mean?
After a quiet period (15 minutes by default) we stop the GPU. The next request wakes it; until it is ready the endpoint answers 503 with a Retry-After header. A cold start takes several minutes: a fresh GPU host loads the server and the weights, and longer if they must be downloaded again. Great for batch jobs and prototypes; pick “always on” for live traffic.
EU or Global?
EU — hosting in EU, built for GDPR, aligned with the EU AI Act. Runs on GPUs of an EU provider in EU data centres. Global — lowest price. GPUs from independent and data-centre providers worldwide. Choose EU for personal or regulated data.
Do you store my prompts?
No. Requests pass straight through to your model server; we keep only counts (requests, tokens, errors) for your dashboard. Your prompts are processed on the GPU that runs your endpoint — on Global that GPU belongs to an independent or data-centre provider, so choose EU for personal or regulated data.
