Grillcheese Research Laboratory
An independent AI research lab in Lévis, Québec.
It all started in a room with nothing left to lose
Grillcheese didn't begin with a grant or a cluster of GPUs. Back in 2023, when this field was brand new, it began with one person, a single consumer graphics card — an AMD Vega 64, 8 GB of memory — and no budget to rent anything bigger. Running modern AI on that was supposed to be impossible, and that constraint turned out to be the whole point: when you can't buy your way around a problem, you're forced to actually solve it. The efficiency this lab is built on didn't come from a strategy deck. It came from necessity. It's also why we wrote our own GPU framework rather than pay the CUDA tax — Grilly, which now runs on almost any card and is free for anyone to use.
We're small, independent, and not attached to anyone's roadmap but our own — because there's no one else's money in the room. Everything we've built since 2023 has been entirely self-funded: no venture round, no outside capital, nothing. That independence isn't a hardship to apologise for; it's the proof. If this work truly needed billions, it wouldn't exist.
Why "Grillcheese"
The name isn't a joke we got stuck with — it's the point. A grilled cheese is comfort food: warm, familiar, the opposite of intimidating. So much of AI today arrives as something to be afraid of — too new, too capable, too opaque, aimed at people rather than made for them. We wanted to stand for the other feeling: the reassurance that something better is coming, and that it's being built with people in mind first. Comfort, not fear. That's the name, and it's the whole intention behind the work.
Being a francophone lab shaped it more than you might expect. Cubby is bilingual from the ground up, English and French treated as first-class rather than one bolted onto the other, because that's the world we actually build and think in. A tool that only works well in one language was never going to be the tool we wanted.
What ties all of it together is one aim: AI that's sustainable, open, and yours to manage — greener than what we have today, and we'll never hide the compute it still takes to get there.
We'd rather be judged on what we can show than on what we can announce. So this site says plainly what's running, what's still in training, and what's only a plan for now — and we'll tell you which is which.