Skip to content

Knowledge Graph

Rook is backed by a knowledge graph that remembers. It’s why Rook can resolve your intent to the right command or component so quickly — and why it gets better over time.

  • 197 Rhino commands — with 543 logged observations of what works and what fails
  • 945 Grasshopper components — mapped to 1,533 intents

When you describe what you want, Rook consults this graph to find the right pieces and how they fit together.

The graph improves from failure, not routine success. When a mistake gets fixed, correction detection records the learning — so the next agent, and the one after, won’t repeat it.

Rook pulls only as much knowledge as a task needs, keeping responses fast and focused:

TierCostWhen
quick~20 tokensfast check
context~50 tokensroutine use
errors~30 tokenson failure
raw500+ tokensforensic / deep dive

Pulling the right tier instead of everything is roughly a 97% token reduction versus reading raw patterns.

You get the benefit of accumulated expertise — the patterns experienced users rely on — without having to learn or maintain any of it yourself.