An applied lab, not a model shop.
Meredic is a small AI research lab. We work on the structure a model reasons inside, and we ship the products that structure makes possible because the guarantees only hold when both are built by the same people.
What we are working toward
Not the largest model, the one people can actually work alongside. That means an interaction contract you can predict: it proposes, you confirm, it logs what it did and lets you undo it. We would rather be the system others copy the manners of.
More parameters buy fluency, not judgement. We are betting that a better substrate, one that keeps affect and salience as first-class state, beats a larger model on the problems that take days rather than turns. If we are right, a small team can hold its own against far larger ones.
Research that never reaches a product is a paper trail. Cura is the test: if the architecture is any good, it should show up as a calmer morning for someone who has never read a word we published.
How we work
Substrate, model and product are evaluated together. We do not hand a research result over a wall and hope.
Architecture notes and benchmark methodology, so the numbers can be argued with. Open foundational work runs through our sister lab, Latent ↗.
Research stage means research stage. Every programme on this site carries its real status, including the ones that are years out.
A team that fits around one table, choosing depth over surface area. It is the only way the first three commitments survive a deadline.
Come build it with us.
We are small enough that you would know everyone by the end of your first week, and every note that arrives gets read by one of us. If this is the sort of problem you would turn over on a walk anyway, say hello. What you have been building is more useful to us than a CV.