---
title: README
emoji: 🛡
colorFrom: indigo
colorTo: gray
sdk: static
pinned: false
short_description: Own the model. Own the cloud it runs on.
thumbnail: https://continker.ai/og-image.webp
---

<p align="center">
  <img src="https://continker.ai/og-image.webp" alt="Continker" style="max-width: 60%; height: auto;">
</p>

<h1 align="center">Own the model. Own the cloud it runs on.</h1>

<p align="center">Continker is an Amsterdam studio building sovereign AI cloud platforms, applied AI, and local models that run offline. Infrastructure organizations own and run themselves, on European terms.</p>

<div class="grid lg:grid-cols-3 gap-x-4 gap-y-4">
  <div class="border rounded-lg p-4">
    <h3 class="font-semibold mb-1">Sovereign AI cloud platforms</h3>
    <p>The infrastructure layer organizations own and run themselves. Model serving, retrieval, and agent workflows on hardware the customer controls, with audit and air-gap built in.</p>
  </div>
  <div class="border rounded-lg p-4">
    <h3 class="font-semibold mb-1">Applied AI</h3>
    <p>The systems that run on it. Language-model agents, retrieval, and automation built for real operational work.</p>
  </div>
  <div class="border rounded-lg p-4">
    <h3 class="font-semibold mb-1">Local models</h3>
    <p>Small, fine-tuned models that run offline on commodity hardware, so the model layer is owned too.</p>
  </div>
</div>

Sovereignty here is technical, not geographic. It is about who controls the keys, the runtime, and the operators, not where the data sits.

---

## Open work

MetroLLM-Bench: a benchmark for running a transit kiosk from a prose prompt, with open-weight students that handle the task offline on commodity hardware. Published as [arXiv:2609.10016](https://huggingface.co/papers/2609.10016) (September 2026). The 2.6 GB student matches GPT-5.4 at maximum reasoning effort on the deterministic tier and exceeds both GPT-5.6 tiers.

| Model | Size | Runs on |
|-------|-----:|---------|
| [Qwen3.5-2B-metro-v24](https://huggingface.co/continker/Qwen3.5-2B-metro-v24) | 1.2 GB | anything, including CPU-only |
| [Qwen3.5-4B-metro-v24](https://huggingface.co/continker/Qwen3.5-4B-metro-v24) | 2.6 GB | a 16 GB laptop |
| [Qwen3.5-9B-metro-v24](https://huggingface.co/continker/Qwen3.5-9B-metro-v24) | 5.3 GB | a 24 GB machine |
| [Qwen3.5-27B-metro-v24](https://huggingface.co/continker/Qwen3.5-27B-metro-v24) | 16 GB | a 32 GB machine |

[Paper](https://huggingface.co/papers/2609.10016) · [Dataset (955 cases)](https://huggingface.co/datasets/continker/metrollm-bench) · [Live demo](https://huggingface.co/spaces/remcohendriks/metrollm) · [Collection](https://huggingface.co/collections/continker/metrollm-bench-v24-6a35b586a11068e1b1ba3d47) · [Benchmark and code](https://github.com/continker/metrollm-bench)

---

More at [continker.ai](https://continker.ai).