An open-source standard for agentic context
The problem
Ask an AI AGENT "what's our MRR this month?" and it has to guess about metrics, schemas, and lineage. Which table has MRR? Use gross_revenue or net_revenue? How do you filter to this month? It writes SQL that looks right and returns a number that's wrong.
The metric definitions and the rules your analysts follow live in dbt, in Looker, or in another semantic layer. Agents Schema is the standard for making these rules accessible to agents.
What it is
Agents Schema is a set of metadata tables inside your warehouse. It gives agents a standard place to look up the context they need before querying your business data.
Providers write metadata into AGENTS.* from the tools that already define your data, such as:
- dbt — models, columns, descriptions, tests, lineage, and semantic-layer metrics
- BI tools / OSI — governed metrics, datasets, and the SQL patterns used to compute them
- Fivetran — connector, table, column, and sync metadata
An agent can read this metadata with SQL, then use the definitions it finds to query your business tables correctly.
Example
Point the connectors at your dbt project and semantic models. They read your dbt manifest and definitions, write them into agents.*`, and re-run whenever your project changes so the metadata stays current.