If you’re EU based and need a GDPR compliant LLM router for legal (or ethical) reasons, don’t sleep on Cortecs.
(What is an LLM router? It’s a software layer positioned between an application and multiple large language model (LLM) providers that directs incoming prompts to the most suitable model for a specific task. So you get one API URL and specify the model in your calls, whether that model is from Google, DeepSeek, or Mistral. Behind the scenes, the LLM router connects your request with a suitable provider.)
Don’t get me wrong, I really like the OG OpenRouter — IMHO it’s still without equal when it comes to both the number of supported providers and configurability. But as a European business that wants to reduce US-based risks in its supply chain, it might not be a good-enough option: even if you use an EU-based inference provider through OpenRouter, OR itself remains a venture in US jurisdiction.
Based in Vienna (Austria), Cortecs offers a good (small) selection of models, zero data retention (ZDR), and “sovereign cloud” toggles. It’s clearly early days, but good stuff already.
My current go-to example: I like DeepSeek V4 Flash, a very fine offering from the eponymous Chinese lab. With Cortecs, I can do that: with ZDR, fully EU-based, in a 100% GDPR compliant way. One of their supported providers is Tensorix from Ireland, and they offer that model. These days, I get 350 tps throughput with ~1.5s latency, which isn’t bad at all.
BTW, a very nice side effect of using LLM routers: I only have to deal with one party and only get one invoice, even if I use all of their supported providers. But that doesn’t mean lock-in! Cortecs' API is OpenAI-compatible, so if I become unhappy with them at some point, I can just up and leave. Because that API is basically a standard by now, changing providers usually just means changing the base URL and model ID in my tool’s config.
And since I use pi quite a lot, I released npm:pi-cortecs a few days ago. Nothing special — it just adds Cortecs provider support for /login, a new /cortecs-models command, and auto-fetching of their available models. Useful! Here’s the npm package.
So, Cortecs is good. But it’s not the only kid on the block, mind. There’s also Nebius Token Factory, from the Netherlands, which is also good — they offer another set of models, and also support prepaid API billing. Don’t sleep on them, either. 😉
#DataSovereignty #EU #LLM
