A drop-in for the OpenAI SDK
Keep your code. Change two lines and your existing client talks to open models.
Prompts and answers stay in memory only while the request runs.
Open-weight models on GPUs we run. Models also listed on OpenRouter keep the same ids.
Keep your code. Change two lines and your existing client talks to open models.
Prompts and completions are streamed through and never written to disk. We keep token counts for billing, nothing else.
Each response carries its price in USD, so you can attribute spend per feature or user.
Give each key a monthly cap. When it is reached, that key stops, the rest keep running.
A VAT invoice is issued for every top-up and sent to your inbox.
See what every key costs, how fast the model answers and how close you are to a limit. Set a monthly cap and sleep well.
Pay by card from $5. Credit lands in your balance the moment the payment clears.
One key per app or environment, each with an optional monthly budget.
Every request is deducted at the published price. Credit never expires.
The questions your legal and security teams ask, answered before they ask. A Polish company, GPUs in the European Economic Area, nothing stored, and the paperwork ready to download.
MADD, Olsztyn, Poland. Contract under Polish law, Polish VAT invoices sent to KSeF, reverse charge for EU businesses.
Models run on GPUs in Poland, Sweden and Iceland, all inside the European Economic Area. The country of every model is shown on the models page and published in /v1/models.
Prompts and answers live only in memory while the response is generated. No content logs, no training, no switch to forget.
Everything procurement asks for, in Polish and English.
Short answers to what developers and procurement ask most. Anything missing? We answer on business days, usually the same day.
Sign in with email or Google, top up, copy your key.