Hundun builds intelligence for work that does not fit inside a prompt.

We are building generative systems for environments where knowledge is fragmented across documents, decisions, tools, and the quiet assumptions that rarely make it into a clean dataset.

The first surface touches legal work, but the deeper problem is more general: long context that changes meaning as it moves, private knowledge that has to stay inside the boundary, and agents that need to act without turning capability into noise.

The product is designed to operate close to the material and the people responsible for it, so that generation is not separated from review and models can learn from the structure of the work without turning private knowledge into public exhaust.

Our path is to make intelligence less like an answer box and more like an instrument for long-horizon work, where the system can build context, use tools, surface disagreement, and leave enough of a trail for careful people to keep moving.