Morrigan. The modular AI built for local power.

Large language models changed everything, but their limits are structural. They learn nothing after their training: every conversation starts over from the same point. Their size requires datacenters, so the cloud, so dependency. And their compute appetite grows faster than their intelligence.

Waiting for these flaws to vanish with the next version means mistaking the improvement of an architecture for going beyond it.

Morrigan explores another path: a modular intelligence, made of specialised building blocks that cooperate, rather than a monolithic block that does everything on average. The goal is simple to state and hard to reach: the most intelligence per watt, locally. An AI that runs on a machine you own, that learns from its context instead of forgetting it, and that asks no one for permission to work.

We publish the thesis and the milestones, not the manufacturing plans. Research moves faster without handing the recipe to those who already own the datacenters.

Morrigan is in active research: the foundations for indexing and multi-language understanding work on real corpora, and each step is validated by tests before the next one. No announced date, no inflated demo: the notable advances are published in the journal.