The work behind the work.
Eldric ships engineering, but the architecture rests on research — efficient recurrent models, an experimental self-improving module, a long-term memory layer, and distillation that brings large models down to hardware you can afford.
Evidence-grounded logic · data-driven reasoning · intelligence you can cite
Efficient by research, not by brute force.
xLSTM
A modern, efficient successor to the LSTM architecture — chosen so capable AI runs on your own hardware instead of a rented cloud endpoint.
NOVA (experimental)
An optional, experimental module exploring self-improving agents. Off by default and clearly marked as research.
Long-term memory
A memory layer that lets Eldric keep and recall what it has learned, across sessions and over time.
Distillation
Making large models compact enough to run on modest hardware — capability without a data-centre of GPUs.
What we're honest about not having.
Some of this is production, some is experimental, and some is a direction rather than a finished result. We label which is which, and we'd rather you verify it than take a claim on faith. The self-improving work in particular is research — powerful to explore, not something we oversell.
Part of the European AI community.
Eldric is an independent, Vienna-built platform. We integrate open research and contribute engineering — an AI you own, grounded in work you can look up.