Skip to content
Catalyst Docs

Data-source guides

A real warehouse can have thousands of cryptically-named tables — APR_STUDENT_COHORT, ARMS_ATHLETE, dated *_20241121 snapshot copies. No amount of inspecting columns tells the model that ARMS_ATHLETE is the athletics roster, or that the dated copies are throwaways. That’s domain knowledge a human writes down — and in Catalyst, you write it down as a guide.

A guide is a skill attached to a data source that teaches the model what its tables mean. What makes it different from a normal skill:

  • It’s hidden from general chat. A guide never appears in an unrelated conversation, so it doesn’t clutter every chat and your institutional table map doesn’t leak. It still shows up in your Skills list — “hidden” only means it’s not advertised to every conversation.
  • It surfaces only when the AI is working with that database. When the model lists your connections, the guide rides along as a pointer; the model loads it before writing queries against that source. This is the answer to “the AI doesn’t know which of my 5,000 warehouse tables to use.”

Managing guides

Section titled “Managing guides”

From a connection’s Guides section (in the IDE’s Properties pane), you can:

  • Attach an existing skill — pick one of your skills to act as this database’s guide.
  • Create a guide — Catalyst makes a fresh hidden skill named after the source, attaches it, and jumps you to the skill editor to write the body.

A source can have several guides, and a guide can apply to more than one source.