About VipAI
VipAI is an AI API gateway. It puts one key, one endpoint and one bill in front of every major model provider, so an agent or an app can call GPT, Claude, Gemini, DeepSeek and others without five sets of credentials and five invoices.
What it does
A request arrives at api.vipai.site/v1 in one of the protocols the providers themselves speak — OpenAI Chat Completions, OpenAI Responses, Anthropic Messages or Gemini native. The router forwards it to the provider that serves the model you asked for, meters the tokens, and records what it costs. The response is the provider's own, unmodified, so official SDKs and cache behaviour keep working.
We do not translate protocols silently and we do not swap models. If a lane is offline, the request fails loudly instead of quietly answering with something cheaper than you asked for.
Pricing
Billing is per million tokens, priced in USD, with input and output metered separately. There is no subscription: you buy credits, they do not expire, and a balance that runs out stops requests rather than becoming a debt.
Rates sit below the providers' published list prices because we commit to enterprise-scale volume and pass the difference through. The discount is not something you have to take on trust — the pricing page shows the provider's list price beside ours for every model.
Where it runs
The gateway runs on dedicated infrastructure we operate, behind a single encrypted entry point, with the relay, the management API and the website separated by hostname. Browsing the site never sends your credentials to a third party, and the API is reached with a bearer key you can revoke at any time from the dashboard.
Your data
Prompts and completions are processed only to route the request, meter it, bill it, prevent abuse and answer support questions. They are not used to train models and are not sold. The privacy policy has the detail.
Talk to us
Support runs through Telegram at @vipai. Include the request id when something behaves unexpectedly — it is the fastest way to find the call in the logs. If a request turns out not to have been served by the model you selected, the charge is refunded and credited back many times over; the terms are in the terms of service.