Your company's own private AI model
We fine-tune open-source models on your data and run them on a private cloud, behind one OpenAI-compatible API.
Built for one company at a time
Trained on your data
Your model learns your terms, your tone and your formats.
Private by design
Each company gets its own model, its own storage and its own API keys.
Easy to connect
Works with the OpenAI API format, so existing tools connect by changing the address.
Built on leading open-source models
We help you choose the model that fits your task and budget.
The math behind your model
Three steps turn an open-source model into yours.
Adapt
Low-rank updates (LoRA) teach the model your domain without retraining every weight.
Learn
Gradient descent lowers the error step by step on your data.
Compress
Quantization stores weights in fewer bits, so the model needs less memory.
The beauty of the math
Attention is the equation at the heart of modern language models.
Euler's identity
Normal distribution
Fourier series
Everything you need to run it
Local languages
Train your model in the languages your team and customers use.
Key controls
Limit each API key by model and IP address.
Usage tracking
See usage by key and set quotas in your console.
Rate limits
Set limits per user to keep usage under control.
Streaming
Show replies as they are generated.
Dedicated option
Run on a separate GPU instance when you need full isolation.
How it works
Share your data
Send us your documents, questions and answers. We prepare them for training.
We train and test
We fine-tune the model, then test it before it goes live.
Start using it
You get a private API key. Manage keys and usage in your console.
One API call
curl https://api.metis.ing/v1/chat/completions \
-H "Authorization: Bearer $METIS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "your-company-model",
"messages": [{"role": "user", "content": "Hello"}]
}'