Querygraph is an AI Navigator over governed enterprise data. It starts from a simple disagreement with the dominant direction of AI infrastructure: serious knowledge work should not require throwing enormous, global, poorly scoped contexts at expensive GPU systems whenever a user asks a precise question. Most enterprise and scientific questions are local, contextual, permissioned, and reproducible. They deserve an architecture built for that reality.
Querygraph treats a lakehouse as more than tables: each dataset is described by Semantic Croissant, projected through CDIF, governed by RBAC and ODRL, addressed by DIDs, and audited through OpenLineage. The result is a focused retrieval and reasoning environment where agents operate inside precise semantic contexts rather than wandering through giant prompt buffers.
The implementation in this repository is intentionally practical. It can load Dataverse and CODATA data into Sail, materialize typed tables, generate Croissant and CDIF sidecars, wrap agent requests with TypeSec TypeDID envelopes, and emit OpenLineage events back into Sail. The point of this book is to make that architecture legible as a textbook: first the motivation, then the pieces, then the full working system.
The system rests on three coordinated, named open-source releases that this book tracks throughout: Grust 0.12.1 “Lobster” for the graph and query substrate, TypeSec 0.13.1 “Torcello” for the typed security fabric, and LakeCat 0.3.0 “Ocelot” for the catalog boundary. Where a chapter leans on a specific capability, it names the release that brought it — many foundations arrived with the previous line (Grust “Crab”, TypeSec “Burano”, LakeCat “Lynx”), and the chapters keep that history.