Product releases, what we build under the hood, and how we work on EU-sovereign AI infrastructure.
A clear table with masked keys and usage per key, plus BYOK for Autopilot and a new MCP connector.
Now with automatic model choice (Autopilot) and insight into which provider handled your traffic.
The default model on chat.loes.ai now runs a LoRA finetune on top of Qwen3.5-27B, dedicated on a single H100.
A test environment for people with their own GPU: sign up as a tester and run along with what we build.
Anyone in Europe with a modern NVIDIA card can install our agent and earn — from €0.04/hour for an RTX 3060 up to €1.57/hour for an H100.
An interview about building Dutch AI infrastructure in Groningen.
Not a finetune of somebody else’s model but an architecture of our own, pretrained by us. Now an experimental preview in Loes Lab.
Jantine Doornbos argues the Netherlands needs an AI assistant of its own — one we control, and one that fits the local culture. It became Loes.ai.
An automatic top-up as soon as your balance drops below a threshold you set, with safety limits built in.
Connect your Mac or PC with a GPU to HostYourAI. You help train Loes — never customer traffic — and get paid per second.
Where Loes answers fast, Super Loes takes its time to reason. Both sit in the model picker by default.
HostYourAI and Loes are growing fast — we’re hiring across engineering, model quality, inference, frontend and sales.
On Loes as a Dutch alternative to ChatGPT.
A Dutch alternative to ChatGPT that runs locally and safely.
AG Connect on Loes, the Dutch-speaking AI chat assistant.
A Groningen founder builds the first Dutch AI assistant.
2.8 trillion parameters and a 1M token context window, on 8 B300s in a European datacenter. Your data never leaves the EU.
On De Grote Tech Show the hosts call Loes AI the Dutch version of ChatGPT.