Vi startet Kapllan fordi vi ville gjøre nøye arbeid i et hastverkets felt.
Kunstig intelligens har i 2026 karakteren av en brannøvelse. Hvert kvartal kommer en ny evne, et nytt referansemål, en ny bekymring. Det er lett å forveksle bevegelse med fremgang, og enda vanskeligere å motstå følelsen av at man må levere noe — hva som helst — før uken er omme.
Vi mener feltet er viktig nok til at dette er feil reaksjon. Systemene som bygges nå vil leve i klasserom, klinikker og domstoler i tiår. De fortjener å bygges på den langsomme måten: med begripelige metoder, ærlige evalueringer og en vilje til å ta feil offentlig.
Kapllan er et laboratorium, ikke en plattform. Vår flaggskipsmodell Llana er resultatet av fire års tålmodig uenighet blant et lite antall forskere. Vi publiserer antagelsene våre. Vi publiserer tilfellene der antagelsene ikke holder. Vi publiserer bitene som ikke fungerte.
Vi forsøker ikke å bygge generell intelligens. Vi forsøker å bygge en bestemt form for intelligens — en som resonnerer begripelig, avslår kalibrert og forblir nyttig i kanten av sin kompetanse. Det er mindre ambisiøst enn noen av våre kolleger. Vi mener det er mer ærlig.
A lab born from the belief that intelligence should be owned, not borrowed.
The first thesis.
NxAI begins as an exploration of sovereign intelligence. The initial idea is simple but foundational: intelligence should be owned, not borrowed. What starts as experiments around agents, private AI systems, memory and reasoning becomes the beginning of a broader lab direction.
Lab mode.
NxAI enters lab mode. Over several months, the work moves from concept to working systems: local infrastructure, agent architecture, voice experiments, RAG, memory, orchestration and early model pipelines. The focus shifts from building “an app” to building AI capability.
Infrastructure becomes real.
The private compute foundation is established. Kapllan / NxAI begins operating its own DGX Spark cluster and orchestration layer. Ray, vLLM, Qdrant, MinIO, FastAPI and internal services become the foundation for a self-owned AI stack.
One brain, many systems.
The architecture expands beyond a single product. The system begins to take shape as a family of connected AI capabilities: Pllan, Llana, Kapllan, Boardroom, Voice, NxNote and Ofelia. The direction becomes clearer — one shared intelligence layer, many interfaces and products.
Brainfleet and Boardroom.
Reasoning becomes a system. Boardroom is introduced as a multi-agent reasoning layer, while NxNote becomes the structured memory layer. Together they form the early idea of Brainfleet: a system for decision intelligence, knowledge reuse and complex reasoning.
From NxAI to Kapllan AI.
The lab receives its name. NxAI evolves into Kapllan AI — a Nordic AI lab built to develop, preserve and transform AI competence into long-term capability. The name Kapllan is inspired by Kapllan Hasani, originating from Kosovo, and carries the meaning and spirit of the knight in Albanian: courage, protection, independence and duty.
The model path begins.
K—1 becomes the first foundation-model program. Kapllan AI starts defining its own model ladder: K—0, K—1, K—2 and beyond. K—1 is the first serious step toward a Nordic, sovereign foundation-model capability — not as a borrowed wrapper, but as an owned model path.
From project to lab.
Kapllan AI is not built around one app, one agent or one customer wedge. It is built around the long-term ambition to create sovereign AI capability — infrastructure, models, agents, memory, reasoning and human-facing systems owned and developed from within.
A small team, deliberately. Hiring slowly.
Fiola Hassani
A founder, an advisor, and a careful idea. We are looking for the next few.
Researchers, engineers, and operators who would rather be right than fast. Remote-friendly, with hubs in Stockholm and Oslo.
See open roles →