Retrieval for a world of agents

To whoever is still reading,

Welcome to Quernstone. We are a small applied lab building the retrieval layer the coming web will stand on.

We spent six years on the unglamorous parts: crawling, deduplication, and embeddings that survive a rewrite. The readers we were building for turned out not to be people.

Today Quernstone answers about 1.4 billion queries a month for teams at Ridgemoss, Larkmoor, Blythewick and Emberline, and for roughly 210,000 independent builders.

Within a decade the volume of machine-issued lookups will dwarf everything humans have ever typed into a box. That shift decides who gets to describe the world, and to whom.

What a machine can find sets the ceiling on what it can be trusted to say, and that ceiling is now load-bearing for a lot of ordinary life. Rumours travel faster than corrections, and a model with a thin index will repeat both with the same confidence. Get the retrieval right and everything downstream inherits it.