Sovereign AI Inference for Europe: Polarise Drive
Every AI feature runs on the same thing underneath: inference. Someone has to actually run the model when a user sends a request. Today, most of that happens on infrastructure outside Europe, even when the product itself is built by a European company.
Polarise Drive changes where that inference happens, without changing how you build.
What Polarise Drive Is
Polarise Drive is a managed inference platform built on two components: a Model Gateway for standardized access to AI models, and an Agentic Engine for building and running AI agents (upcoming). Both work standalone, and both work together.
The Model Gateway is OpenAI SDK compatible. If your product already talks to an OpenAI-compatible API, you don’t rewrite anything. You change the base URL and the API key, and requests start running on Polarise infrastructure instead.
Sovereign AI Inference, in Practice
“Sovereign AI” gets used loosely. In practice, it means one thing: your data is processed under a legal framework you actually control. Polarise Drive runs on European infrastructure, operated by a European legal entity, under EU law. That means: Polarise owns the full stack from infrastructure to token, without third party risk.
This matters for any company handling customer data, regulated data, or anything falling under the EU AI Act, where the question isn’t just “is my provider fast enough,” it’s “who can access this data, and under which jurisdiction.”
Models Built for Different Jobs
Drive gives you access to a growing collection of open-weight models.
Available today:
| Model | Provider | Task |
|---|---|---|
| GLM_5.2 | Zhipu AI | Text Generation (general-purpose) |
| Qwen_3.8-27b | Qwen | Text Generation |
| Nemotron_3-ultra | Nvidia | Text Generation |
| Deepseek_v4-flash | DeepSeek | Text Generation |
Upcoming:
| Model | Provider | Task |
|---|---|---|
| Mistral Large 3 | Mistral AI | Text Generation |
| Mistral Medium 3.5 | Mistral AI | Text Generation |
| Kimi K2.7 Code | Moonshot AI | Code Generation |
| gemma-4-31b | Vision / Image-to-Text | |
| DeepSeek-OCR | DeepSeek AI | OCR / Document Understanding |
| EmbeddingGemma-300M | Embeddings | |
| Qwen3-Reranker-4B | Qwen | Reranking |
| Qwen3-ASR | Qwen | Speech-to-Text |
| GLM-ASR-Nano | Zhipu AI | Speech-to-Text |
New models are added monthly. Don’t see what you need? to reach out and we’ll add it.
Getting Started
There’s no migration project. You create an account, generate an API key, and point your existing integration at the new endpoint. If you’re already using an OpenAI-compatible setup, that’s the entire change.
import openai
import os
client = openai.OpenAI(
api_key=os.environ.get("POLARISE_API_KEY"),
base_url='https://api.drive.polarise.com/v1'
)
completion = client.chat.completions.create(
messages=[{
'role': 'user',
'content': 'Summarise this contract clause.'
}],
model='glm-5.2'
)Want to see how fast this actually is? to request access or explore documentation.

