ChatGPT for your company, on hardware only you use
A GPU server with Qwen3 served through an OpenAI-compatible API and a web chat for the team, installed at first boot. Documents and prompts never leave the server. Point every tool you use at it with one URL and one key.
From €312.38 per month (€0.4339 per hour) on the AI-RTX2000 plan, billed hourly.
Dedicated GPU
NVIDIA RTX 2000 or 4000 Ada with 16 or 20 GB of VRAM, not shared with anyone. Predictable latency at any hour.
OpenAI-compatible API
Any SDK, agent or tool that speaks the OpenAI API works unchanged: change the base URL and the key, nothing else.
Chat for the team
A web interface with accounts, chat history and document upload, served over HTTPS on your server's address.
One click to use everywhere
From the server page, make it your account's AI endpoint; every agent and AI app you order afterwards talks to your own model.
How it works
The model is on your GPU a few minutes after the order.
- 1
Order a GPU plan with Private AI
The install sets up the drivers, the inference server and the chat UI, and generates your API key.
- 2
Wait for the model download
The first boot pulls an 8-billion-parameter model. The API answers a few minutes later; the key is on your server page.
- 3
Connect your tools
Use the API URL and key in your apps, or press Use as my AI endpoint and order agents that use it automatically.
Questions
- Which model runs?
- Qwen3-8B, an open-weight model under the Apache 2.0 licence, quantised to fit the GPU with room for an 8k context. You can switch to another model from the server with one variable.
- Is it GDPR-safe?
- Prompts, documents and answers stay on a server in Romania that only you can reach, with no third-party API involved. You remain the controller of the data.
- How fast is it?
- Tens of tokens per second for a single user on the 16 GB card, enough for chat, summaries, RAG and agents. Batch workloads scale with the dual-GPU plans.
- What does it cost?
- The GPU plan, billed by the hour up to its monthly price. Stop the server when it is not needed and pay only for storage.
Put your own model to work
Order from the console on a GPU plan. The model downloads on first boot and answers a few minutes later.
Create a Private AI server