Introducing Guides: Warehouse-Native Context for AI Agents

​If you've pointed an AI agent at a data warehouse, you know how this goes. It digs through schemas, guesses at join relationships, and comes back with an answer that sounds confident and reads well. It's also frequently wrong. The bigger your warehouse, the worse it gets.

​Guides are MotherDuck's fix: a context layer that lives in the warehouse itself, versioned and queryable like any other object. The results surprised even us. On the DABStep benchmark, Guides raised a small, fast model's accuracy by 72 percentage points and cut costs by 55%. That's a bigger gain than swapping in the latest frontier model. The MotherDuck team will show you how it works, live.

​You'll learn how to:

• ​Bootstrap your first Guides from context your team already has

• ​Keep metric definitions consistent with org-level Guides that update everywhere at once

• ​Manage Guides in plain SQL, so version control and CI just work

• ​Use Guides from Claude, Codex, Cursor, or any MCP-enabled agent